Data Preprocessing : Remedial Procedures for Skewed Data with Extreme Outliers

1. Table of Contents

This project explores different remedial procedures for handling skewed data with extreme outliers for classification using various helpful packages in R. Using a Logistic Regression model structure, methods applied in the analysis to address data distribution skewness and outlying points included the Box-Cox Transformation, Yeo-Johnson Transformation, Exponential Transformation, Inverse Hyperbolic Sine Transformation, Base-10 Logarithm Transformation, Natural Logarithm Transformation, Square Root Transformation, Outlier Winsorization Treatment and Outlier Spatial Sign Treatment. The resulting predictions derived from the candidate models applying various remedial procedures were evaluated in terms of their discrimination power using the area under the receiver operating characteristic curve (AUROC) metric. The AUROC values were compared to that of the baseline model which made use of data without any form of data transformation and treatment. All results were consolidated in a Summary presented at the end of the document.

Data transformation is the application of mathematical modifications to the values of a variable which changes and improves its distributional characteristics. A transformed scale may be used to reduce distributional skewness and produce approximately equal spreads minimizing outliers which better support standardized model fit indices and effect size estimates assumed by most parametric statistical tests and modelling procedures. The methods applied in this study (mostly contained in the stats and caret package) attempt to explore various data transformation strategies for non-normal distributions in readying the data for analysis procedures requiring improved characteristics of normality, homogeneity of variance and additivity of relationships.

1.1 Sample Data

The Solubility dataset from the AppliedPredictiveModeling package was used for this illustrated example. The original numeric response was transformed to simulate a dichotomous categorical variable. Other original predictors were removed from the dataset leaving only a subset of numeric predictors used during the analysis.

Preliminary dataset assessment:

[A] 1267 rows (observations)
     [A.1] Train Set = 951 observations
     [A.2] Test Set = 316 observations

[B] 5 columns (variables)
     [B.1] 1/5 response = Log_Solubility_Class variable (factor)
            [B.1.1] Levels = Log_Solubility_Class=Low < Log_Solubility_Class=High
     [B.2] 4/5 predictors = All remaining variables (4/4 numeric)

Code Chunk | Output

##################################
# Loading R libraries
##################################
library(AppliedPredictiveModeling)
library(caret)
library(rpart)
library(lattice)
library(dplyr)
library(tidyr)
library(moments)
library(skimr)
library(RANN)
library(pls)
library(corrplot)
library(tidyverse)
library(lares)
library(DMwR2)
library(gridExtra)
library(rattle)
library(rpart.plot)
library(RColorBrewer)
library(stats)
library(nnet)
library(elasticnet)
library(earth)
library(party)
library(kernlab)
library(randomForest)
library(Cubist)
library(pROC)
library(mda)
library(klaR)
library(pamr)

##################################
# Loading source and
# formulating the train set
##################################
data(solubility)
Solubility_Train <- as.data.frame(cbind(solTrainY,solTrainX))
Solubility_Test  <- as.data.frame(cbind(solTestY,solTestX))

##################################
# Applying dichotomization and
# defining the response variable
##################################
Solubility_Train$Log_Solubility_Class <- ifelse(Solubility_Train$solTrainY<mean(Solubility_Train$solTrainY),
                                                "Low","High")
Solubility_Train$Log_Solubility_Class <- factor(Solubility_Train$Log_Solubility_Class,
                                                levels = c("Low","High"))
Solubility_Test$Log_Solubility_Class <- ifelse(Solubility_Test$solTestY<mean(Solubility_Train$solTrainY),
                                                "Low","High")
Solubility_Test$Log_Solubility_Class <- factor(Solubility_Test$Log_Solubility_Class,
                                                levels = c("Low","High"))

Solubility_Train$solTrainY <- NULL
Solubility_Test$solTestY <- NULL

##################################
# Filtering in a subset of variables
# for the analysis
##################################
Solubility_Train <- Solubility_Train[,c("HydrophilicFactor",
                                        "NumAtoms",
                                        "NumNonHAtoms",
                                        "NumCarbon",
                                        "Log_Solubility_Class")]

Solubility_Test <- Solubility_Test[,c("HydrophilicFactor",
                                      "NumAtoms",
                                      "NumNonHAtoms",
                                      "NumCarbon",
                                      "Log_Solubility_Class")]

##################################
# Performing a general exploration of the train set
##################################
dim(Solubility_Train)

## [1] 951   5

str(Solubility_Train)

## 'data.frame':    951 obs. of  5 variables:
##  $ HydrophilicFactor   : num  -0.856 -0.37 -0.33 -0.96 -0.069 -0.651 -0.729 -0.835 0.194 0.353 ...
##  $ NumAtoms            : int  28 49 33 26 31 32 35 38 56 37 ...
##  $ NumNonHAtoms        : int  16 26 15 10 15 15 23 14 27 17 ...
##  $ NumCarbon           : int  14 21 13 10 9 10 17 12 22 14 ...
##  $ Log_Solubility_Class: Factor w/ 2 levels "Low","High": 1 1 1 1 1 1 1 1 1 1 ...

summary(Solubility_Train)

##  HydrophilicFactor     NumAtoms      NumNonHAtoms     NumCarbon     
##  Min.   :-0.98500   Min.   : 5.00   Min.   : 2.00   Min.   : 1.000  
##  1st Qu.:-0.76300   1st Qu.:17.00   1st Qu.: 8.00   1st Qu.: 6.000  
##  Median :-0.31400   Median :22.00   Median :12.00   Median : 9.000  
##  Mean   :-0.02059   Mean   :25.51   Mean   :13.16   Mean   : 9.893  
##  3rd Qu.: 0.31300   3rd Qu.:31.00   3rd Qu.:17.00   3rd Qu.:12.000  
##  Max.   :13.48300   Max.   :94.00   Max.   :47.00   Max.   :33.000  
##  Log_Solubility_Class
##  Low :427            
##  High:524            
##                      
##                      
##                      
## 

##################################
# Performing a general exploration of the test set
##################################
dim(Solubility_Test)

## [1] 316   5

str(Solubility_Test)

## 'data.frame':    316 obs. of  5 variables:
##  $ HydrophilicFactor   : num  0.492 1.317 0.846 0.984 0.843 ...
##  $ NumAtoms            : int  8 13 14 19 15 8 8 13 13 17 ...
##  $ NumNonHAtoms        : int  5 6 8 7 9 4 4 5 5 8 ...
##  $ NumCarbon           : int  2 3 6 5 6 2 3 4 4 7 ...
##  $ Log_Solubility_Class: Factor w/ 2 levels "Low","High": 2 2 2 2 2 2 2 2 2 2 ...

summary(Solubility_Test)

##  HydrophilicFactor    NumAtoms     NumNonHAtoms     NumCarbon     
##  Min.   :-0.9860   Min.   : 5.0   Min.   : 3.00   Min.   : 1.000  
##  1st Qu.:-0.7670   1st Qu.:17.0   1st Qu.: 8.00   1st Qu.: 6.000  
##  Median :-0.3970   Median :22.0   Median :11.00   Median : 8.000  
##  Mean   :-0.1022   Mean   :24.6   Mean   :12.71   Mean   : 9.785  
##  3rd Qu.: 0.2140   3rd Qu.:29.0   3rd Qu.:16.00   3rd Qu.:12.000  
##  Max.   : 5.0000   Max.   :68.0   Max.   :33.00   Max.   :24.000  
##  Log_Solubility_Class
##  Low :143            
##  High:173            
##                      
##                      
##                      
## 

##################################
# Formulating a data type assessment summary
##################################
PDA <- Solubility_Train
(PDA.Summary <- data.frame(
  Column.Index=c(1:length(names(PDA))),
  Column.Name= names(PDA), 
  Column.Type=sapply(PDA, function(x) class(x)), 
  row.names=NULL)
)

##   Column.Index          Column.Name Column.Type
## 1            1    HydrophilicFactor     numeric
## 2            2             NumAtoms     integer
## 3            3         NumNonHAtoms     integer
## 4            4            NumCarbon     integer
## 5            5 Log_Solubility_Class      factor

1.2 Data Quality Assessment

[A] No missing observations noted for any variable.

[B] No low variance observed for any variable with First.Second.Mode.Ratio>5.

[C] No low variance observed for any variable with Unique.Count.Ratio<0.01.

[D] High skewness observed for 1 variables with Skewness>3 or Skewness<(-3).
     [D.1] HydrophilicFactor variable (numeric)

Code Chunk | Output

##################################
# Loading dataset
##################################
DQA <- Solubility_Train

##################################
# Formulating an overall data quality assessment summary
##################################
(DQA.Summary <- data.frame(
  Column.Index=c(1:length(names(DQA))),
  Column.Name= names(DQA), 
  Column.Type=sapply(DQA, function(x) class(x)), 
  Row.Count=sapply(DQA, function(x) nrow(DQA)),
  NA.Count=sapply(DQA,function(x)sum(is.na(x))),
  Fill.Rate=sapply(DQA,function(x)format(round((sum(!is.na(x))/nrow(DQA)),3),nsmall=3)),
  row.names=NULL)
)

##   Column.Index          Column.Name Column.Type Row.Count NA.Count Fill.Rate
## 1            1    HydrophilicFactor     numeric       951        0     1.000
## 2            2             NumAtoms     integer       951        0     1.000
## 3            3         NumNonHAtoms     integer       951        0     1.000
## 4            4            NumCarbon     integer       951        0     1.000
## 5            5 Log_Solubility_Class      factor       951        0     1.000

##################################
# Listing all predictors
##################################
DQA.Predictors <- DQA[,!names(DQA) %in% c("Log_Solubility_Class")]

##################################
# Listing all numeric predictors
##################################
DQA.Predictors.Numeric <- DQA.Predictors[,sapply(DQA.Predictors, is.numeric)]

if (length(names(DQA.Predictors.Numeric))>0) {
    print(paste0("There are ",
               (length(names(DQA.Predictors.Numeric))),
               " numeric predictor variable(s)."))
} else {
  print("There are no numeric predictor variables.")
}

## [1] "There are 4 numeric predictor variable(s)."

##################################
# Listing all factor predictors
##################################
DQA.Predictors.Factor <- DQA.Predictors[,sapply(DQA.Predictors, is.factor)]

if (length(names(DQA.Predictors.Factor))>0) {
    print(paste0("There are ",
               (length(names(DQA.Predictors.Factor))),
               " factor predictor variable(s)."))
} else {
  print("There are no factor predictor variables.")
}

## [1] "There are no factor predictor variables."

##################################
# Formulating a data quality assessment summary for factor predictors
##################################
if (length(names(DQA.Predictors.Factor))>0) {
  
  ##################################
  # Formulating a function to determine the first mode
  ##################################
  FirstModes <- function(x) {
    ux <- unique(na.omit(x))
    tab <- tabulate(match(x, ux))
    ux[tab == max(tab)]
  }

  ##################################
  # Formulating a function to determine the second mode
  ##################################
  SecondModes <- function(x) {
    ux <- unique(na.omit(x))
    tab <- tabulate(match(x, ux))
    fm = ux[tab == max(tab)]
    sm = x[!(x %in% fm)]
    usm <- unique(sm)
    tabsm <- tabulate(match(sm, usm))
    ifelse(is.na(usm[tabsm == max(tabsm)])==TRUE,
           return("x"),
           return(usm[tabsm == max(tabsm)]))
  }
  
  (DQA.Predictors.Factor.Summary <- data.frame(
  Column.Name= names(DQA.Predictors.Factor), 
  Column.Type=sapply(DQA.Predictors.Factor, function(x) class(x)), 
  Unique.Count=sapply(DQA.Predictors.Factor, function(x) length(unique(x))),
  First.Mode.Value=sapply(DQA.Predictors.Factor, function(x) as.character(FirstModes(x)[1])),
  Second.Mode.Value=sapply(DQA.Predictors.Factor, function(x) as.character(SecondModes(x)[1])),
  First.Mode.Count=sapply(DQA.Predictors.Factor, function(x) sum(na.omit(x) == FirstModes(x)[1])),
  Second.Mode.Count=sapply(DQA.Predictors.Factor, function(x) sum(na.omit(x) == SecondModes(x)[1])),
  Unique.Count.Ratio=sapply(DQA.Predictors.Factor, function(x) format(round((length(unique(x))/nrow(DQA.Predictors.Factor)),3), nsmall=3)),
  First.Second.Mode.Ratio=sapply(DQA.Predictors.Factor, function(x) format(round((sum(na.omit(x) == FirstModes(x)[1])/sum(na.omit(x) == SecondModes(x)[1])),3), nsmall=3)),
  row.names=NULL)
  )
  
} 

##################################
# Formulating a data quality assessment summary for numeric predictors
##################################
if (length(names(DQA.Predictors.Numeric))>0) {
  
  ##################################
  # Formulating a function to determine the first mode
  ##################################
  FirstModes <- function(x) {
    ux <- unique(na.omit(x))
    tab <- tabulate(match(x, ux))
    ux[tab == max(tab)]
  }

  ##################################
  # Formulating a function to determine the second mode
  ##################################
  SecondModes <- function(x) {
    ux <- unique(na.omit(x))
    tab <- tabulate(match(x, ux))
    fm = ux[tab == max(tab)]
    sm = na.omit(x)[!(na.omit(x) %in% fm)]
    usm <- unique(sm)
    tabsm <- tabulate(match(sm, usm))
    ifelse(is.na(usm[tabsm == max(tabsm)])==TRUE,
           return(0.00001),
           return(usm[tabsm == max(tabsm)]))
  }
  
  (DQA.Predictors.Numeric.Summary <- data.frame(
  Column.Name= names(DQA.Predictors.Numeric), 
  Column.Type=sapply(DQA.Predictors.Numeric, function(x) class(x)), 
  Unique.Count=sapply(DQA.Predictors.Numeric, function(x) length(unique(x))),
  Unique.Count.Ratio=sapply(DQA.Predictors.Numeric, function(x) format(round((length(unique(x))/nrow(DQA.Predictors.Numeric)),3), nsmall=3)),
  First.Mode.Value=sapply(DQA.Predictors.Numeric, function(x) format(round((FirstModes(x)[1]),3),nsmall=3)),
  Second.Mode.Value=sapply(DQA.Predictors.Numeric, function(x) format(round((SecondModes(x)[1]),3),nsmall=3)),
  First.Mode.Count=sapply(DQA.Predictors.Numeric, function(x) sum(na.omit(x) == FirstModes(x)[1])),
  Second.Mode.Count=sapply(DQA.Predictors.Numeric, function(x) sum(na.omit(x) == SecondModes(x)[1])),
  First.Second.Mode.Ratio=sapply(DQA.Predictors.Numeric, function(x) format(round((sum(na.omit(x) == FirstModes(x)[1])/sum(na.omit(x) == SecondModes(x)[1])),3), nsmall=3)),
  Minimum=sapply(DQA.Predictors.Numeric, function(x) format(round(min(x,na.rm = TRUE),3), nsmall=3)),
  Mean=sapply(DQA.Predictors.Numeric, function(x) format(round(mean(x,na.rm = TRUE),3), nsmall=3)),
  Median=sapply(DQA.Predictors.Numeric, function(x) format(round(median(x,na.rm = TRUE),3), nsmall=3)),
  Maximum=sapply(DQA.Predictors.Numeric, function(x) format(round(max(x,na.rm = TRUE),3), nsmall=3)),
  Skewness=sapply(DQA.Predictors.Numeric, function(x) format(round(skewness(x,na.rm = TRUE),3), nsmall=3)),
  Kurtosis=sapply(DQA.Predictors.Numeric, function(x) format(round(kurtosis(x,na.rm = TRUE),3), nsmall=3)),
  Percentile25th=sapply(DQA.Predictors.Numeric, function(x) format(round(quantile(x,probs=0.25,na.rm = TRUE),3), nsmall=3)),
  Percentile75th=sapply(DQA.Predictors.Numeric, function(x) format(round(quantile(x,probs=0.75,na.rm = TRUE),3), nsmall=3)),
  row.names=NULL)
  )  
  
}

##         Column.Name Column.Type Unique.Count Unique.Count.Ratio
## 1 HydrophilicFactor     numeric          369              0.388
## 2          NumAtoms     integer           66              0.069
## 3      NumNonHAtoms     integer           36              0.038
## 4         NumCarbon     integer           28              0.029
##   First.Mode.Value Second.Mode.Value First.Mode.Count Second.Mode.Count
## 1           -0.828            -0.158               21                20
## 2           22.000            24.000               73                51
## 3            8.000            11.000              104                73
## 4            6.000             7.000              105                97
##   First.Second.Mode.Ratio Minimum   Mean Median Maximum Skewness Kurtosis
## 1                   1.050  -0.985 -0.021 -0.314  13.483    3.404   27.504
## 2                   1.431   5.000 25.507 22.000  94.000    1.364    5.523
## 3                   1.425   2.000 13.161 12.000  47.000    0.993    4.129
## 4                   1.082   1.000  9.893  9.000  33.000    0.927    3.616
##   Percentile25th Percentile75th
## 1         -0.763          0.313
## 2         17.000         31.000
## 3          8.000         17.000
## 4          6.000         12.000

##################################
# Identifying potential data quality issues
##################################

##################################
# Checking for missing observations
##################################
if ((nrow(DQA.Summary[DQA.Summary$NA.Count>0,]))>0){
  print(paste0("Missing observations noted for ",
               (nrow(DQA.Summary[DQA.Summary$NA.Count>0,])),
               " variable(s) with NA.Count>0 and Fill.Rate<1.0."))
  DQA.Summary[DQA.Summary$NA.Count>0,]
} else {
  print("No missing observations noted.")
}

## [1] "No missing observations noted."

##################################
# Checking for zero or near-zero variance predictors
##################################
if (length(names(DQA.Predictors.Factor))==0) {
  print("No factor predictors noted.")
} else if (nrow(DQA.Predictors.Factor.Summary[as.numeric(as.character(DQA.Predictors.Factor.Summary$First.Second.Mode.Ratio))>5,])>0){
  print(paste0("Low variance observed for ",
               (nrow(DQA.Predictors.Factor.Summary[as.numeric(as.character(DQA.Predictors.Factor.Summary$First.Second.Mode.Ratio))>5,])),
               " factor variable(s) with First.Second.Mode.Ratio>5."))
  DQA.Predictors.Factor.Summary[as.numeric(as.character(DQA.Predictors.Factor.Summary$First.Second.Mode.Ratio))>5,]
} else {
  print("No low variance factor predictors due to high first-second mode ratio noted.")
}

## [1] "No factor predictors noted."

if (length(names(DQA.Predictors.Numeric))==0) {
  print("No numeric predictors noted.")
} else if (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$First.Second.Mode.Ratio))>5,])>0){
  print(paste0("Low variance observed for ",
               (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$First.Second.Mode.Ratio))>5,])),
               " numeric variable(s) with First.Second.Mode.Ratio>5."))
  DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$First.Second.Mode.Ratio))>5,]
} else {
  print("No low variance numeric predictors due to high first-second mode ratio noted.")
}

## [1] "No low variance numeric predictors due to high first-second mode ratio noted."

if (length(names(DQA.Predictors.Numeric))==0) {
  print("No numeric predictors noted.")
} else if (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Unique.Count.Ratio))<0.01,])>0){
  print(paste0("Low variance observed for ",
               (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Unique.Count.Ratio))<0.01,])),
               " numeric variable(s) with Unique.Count.Ratio<0.01."))
  DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Unique.Count.Ratio))<0.01,]
} else {
  print("No low variance numeric predictors due to low unique count ratio noted.")
}

## [1] "No low variance numeric predictors due to low unique count ratio noted."

##################################
# Checking for skewed predictors
##################################
if (length(names(DQA.Predictors.Numeric))==0) {
  print("No numeric predictors noted.")
} else if (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))>3 |
                                               as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))<(-3),])>0){
  print(paste0("High skewness observed for ",
  (nrow(DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))>3 |
                                               as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))<(-3),])),
  " numeric variable(s) with Skewness>3 or Skewness<(-3)."))
  DQA.Predictors.Numeric.Summary[as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))>3 |
                                 as.numeric(as.character(DQA.Predictors.Numeric.Summary$Skewness))<(-3),]
} else {
  print("No skewed numeric predictors noted.")
}

## [1] "High skewness observed for 1 numeric variable(s) with Skewness>3 or Skewness<(-3)."

##         Column.Name Column.Type Unique.Count Unique.Count.Ratio
## 1 HydrophilicFactor     numeric          369              0.388
##   First.Mode.Value Second.Mode.Value First.Mode.Count Second.Mode.Count
## 1           -0.828            -0.158               21                20
##   First.Second.Mode.Ratio Minimum   Mean Median Maximum Skewness Kurtosis
## 1                   1.050  -0.985 -0.021 -0.314  13.483    3.404   27.504
##   Percentile25th Percentile75th
## 1         -0.763          0.313

1.3 Data Preprocessing

1.3.1 Outlier

[A] Outliers noted for 4 variables with the numeric data visualized through a boxplot including observations classified as suspected outliers using the IQR criterion. The IQR criterion means that all observations above the (75th percentile + 1.5 x IQR) or below the (25th percentile - 1.5 x IQR) are suspected outliers, where IQR is the difference between the third quartile (75th percentile) and first quartile (25th percentile). Outlier treatment for numerical stability remains optional depending on potential model requirements for the subsequent steps.
     [A.1] HydrophilicFactor variable (53 outliers detected)
     [A.2] NumAtoms variable (44 outliers detected)
     [A.3] NumNonHAtoms variable (15 outliers detected)
     [A.4] NumCarbon variable (35 outliers detected)

Code Chunk | Output

##################################
# Loading dataset
##################################
DPA <- Solubility_Train

##################################
# Listing all predictors
##################################
DPA.Predictors <- DPA[,!names(DPA) %in% c("Log_Solubility_Class")]

##################################
# Listing all numeric predictors
##################################
DPA.Predictors.Numeric <- DPA.Predictors[,sapply(DPA.Predictors, is.numeric)]

##################################
# Identifying outliers for the numeric predictors
##################################
OutlierCountList <- c()

for (i in 1:ncol(DPA.Predictors.Numeric)) {
  Outliers <- boxplot.stats(DPA.Predictors.Numeric[,i])$out
  OutlierCount <- length(Outliers)
  OutlierCountList <- append(OutlierCountList,OutlierCount)
  OutlierIndices <- which(DPA.Predictors.Numeric[,i] %in% c(Outliers))
  boxplot(DPA.Predictors.Numeric[,i],
          ylab = names(DPA.Predictors.Numeric)[i],
          main = names(DPA.Predictors.Numeric)[i],
          horizontal=TRUE)
  mtext(paste0(OutlierCount, " Outlier(s) Detected"))
}

OutlierCountSummary <- as.data.frame(cbind(names(DPA.Predictors.Numeric),(OutlierCountList)))
names(OutlierCountSummary) <- c("NumericPredictors","OutlierCount")
OutlierCountSummary$OutlierCount <- as.numeric(as.character(OutlierCountSummary$OutlierCount))
NumericPredictorWithOutlierCount <- nrow(OutlierCountSummary[OutlierCountSummary$OutlierCount>0,])
print(paste0(NumericPredictorWithOutlierCount, " numeric variable(s) were noted with outlier(s)." ))

## [1] "4 numeric variable(s) were noted with outlier(s)."

##################################
# Gathering descriptive statistics
##################################
(DPA_Skimmed <- skim(DPA.Predictors.Numeric))

Data summary

Name	DPA.Predictors.Numeric
Number of rows	951
Number of columns	4
_______________________
Column type frequency:
numeric	4
________________________
Group variables	None

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
HydrophilicFactor	0	1	-0.02	1.13	-0.98	-0.76	-0.31	0.31	13.48	▇▁▁▁▁
NumAtoms	0	1	25.51	12.61	5.00	17.00	22.00	31.00	94.00	▇▆▂▁▁
NumNonHAtoms	0	1	13.16	6.50	2.00	8.00	12.00	17.00	47.00	▇▆▂▁▁
NumCarbon	0	1	9.89	5.29	1.00	6.00	9.00	12.00	33.00	▇▇▃▁▁

###################################
# Verifying the data dimensions
###################################
dim(DPA.Predictors.Numeric)

## [1] 951   4

1.3.2 Zero and Near-Zero Variance

[A] Low variance noted for any variable from the previous data quality assessment using a lower threshold.

[B] No low variance noted for any variables using a preprocessing summary from the caret package. The nearZeroVar method using both the freqCut and uniqueCut criteria set at 95/5 and 10, respectively, were applied on the dataset.

Code Chunk | Output

##################################
# Loading dataset
##################################
DPA <- Solubility_Train

##################################
# Gathering descriptive statistics
##################################
(DPA_Skimmed <- skim(DPA))

Data summary

Name	DPA
Number of rows	951
Number of columns	5
_______________________
Column type frequency:
factor	1
numeric	4
________________________
Group variables	None

Variable type: factor

skim_variable	n_missing	complete_rate	ordered	n_unique	top_counts
Log_Solubility_Class	0	1	FALSE	2	Hig: 524, Low: 427

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
HydrophilicFactor	0	1	-0.02	1.13	-0.98	-0.76	-0.31	0.31	13.48	▇▁▁▁▁
NumAtoms	0	1	25.51	12.61	5.00	17.00	22.00	31.00	94.00	▇▆▂▁▁
NumNonHAtoms	0	1	13.16	6.50	2.00	8.00	12.00	17.00	47.00	▇▆▂▁▁
NumCarbon	0	1	9.89	5.29	1.00	6.00	9.00	12.00	33.00	▇▇▃▁▁

##################################
# Identifying columns with low variance
###################################
DPA_LowVariance <- nearZeroVar(DPA,
                               freqCut = 95/5,
                               uniqueCut = 10,
                               saveMetrics= TRUE)
(DPA_LowVariance[DPA_LowVariance$nzv,])

## [1] freqRatio     percentUnique zeroVar       nzv          
## <0 rows> (or 0-length row.names)

if ((nrow(DPA_LowVariance[DPA_LowVariance$nzv,]))==0){

  print("No low variance predictors noted.")

} else {

  print(paste0("Low variance observed for ",
               (nrow(DPA_LowVariance[DPA_LowVariance$nzv,])),
               " numeric variable(s) with First.Second.Mode.Ratio>4 and Unique.Count.Ratio<0.10."))

  DPA_LowVarianceForRemoval <- (nrow(DPA_LowVariance[DPA_LowVariance$nzv,]))

  print(paste0("Low variance can be resolved by removing ",
               (nrow(DPA_LowVariance[DPA_LowVariance$nzv,])),
               " numeric variable(s)."))

  for (j in 1:DPA_LowVarianceForRemoval) {
  DPA_LowVarianceRemovedVariable <- rownames(DPA_LowVariance[DPA_LowVariance$nzv,])[j]
  print(paste0("Variable ",
               j,
               " for removal: ",
               DPA_LowVarianceRemovedVariable))
  }

  DPA %>%
  skim() %>%
  dplyr::filter(skim_variable %in% rownames(DPA_LowVariance[DPA_LowVariance$nzv,]))

  ##################################
  # Filtering out columns with low variance
  #################################
  DPA_ExcludedLowVariance <- DPA[,!names(DPA) %in% rownames(DPA_LowVariance[DPA_LowVariance$nzv,])]

  ##################################
  # Gathering descriptive statistics
  ##################################
  (DPA_ExcludedLowVariance_Skimmed <- skim(DPA_ExcludedLowVariance))
}

## [1] "No low variance predictors noted."

1.3.3 Collinearity

[A] No high correlation > 95% were noted for any 2 variable pairs as confirmed using the preprocessing summaries from the caret and lares packages.

Code Chunk | Output

##################################
# Loading dataset
##################################
DPA <- Solubility_Train

##################################
# Listing all predictors
##################################
DPA.Predictors <- DPA[,!names(DPA) %in% c("Log_Solubility_Class")]

##################################
# Listing all numeric predictors
##################################
DPA.Predictors.Numeric <- DPA.Predictors[,sapply(DPA.Predictors, is.numeric)]

##################################
# Visualizing pairwise correlation between predictors
##################################
DPA_CorrelationTest <- cor.mtest(DPA.Predictors.Numeric,
                       method = "pearson",
                       conf.level = .95)

corrplot(cor(DPA.Predictors.Numeric,
             method = "pearson",
             use="pairwise.complete.obs"),
         method = "circle",
         type = "upper",
         order = "original",
         tl.col = "black",
         tl.cex = 0.75,
         tl.srt = 90,
         sig.level = 0.05,
         p.mat = DPA_CorrelationTest$p,
         insig = "blank")

##################################
# Identifying the highly correlated variables
##################################
DPA_Correlation <-  cor(DPA.Predictors.Numeric,
                        method = "pearson",
                        use="pairwise.complete.obs")
(DPA_HighlyCorrelatedCount <- sum(abs(DPA_Correlation[upper.tri(DPA_Correlation)]) > 0.95))

## [1] 0

if (DPA_HighlyCorrelatedCount == 0) {
  print("No highly correlated predictors noted.")
} else {
  print(paste0("High correlation observed for ",
               (DPA_HighlyCorrelatedCount),
               " pairs of numeric variable(s) with Correlation.Coefficient>0.95."))

  (DPA_HighlyCorrelatedPairs <- corr_cross(DPA.Predictors.Numeric,
  max_pvalue = 0.05,
  top = DPA_HighlyCorrelatedCount,
  rm.na = TRUE,
  grid = FALSE
))

}

## [1] "No highly correlated predictors noted."

if (DPA_HighlyCorrelatedCount > 0) {
  DPA_HighlyCorrelated <- findCorrelation(DPA_Correlation, cutoff = 0.95)

  (DPA_HighlyCorrelatedForRemoval <- length(DPA_HighlyCorrelated))

  print(paste0("High correlation can be resolved by removing ",
               (DPA_HighlyCorrelatedForRemoval),
               " numeric variable(s)."))

  for (j in 1:DPA_HighlyCorrelatedForRemoval) {
  DPA_HighlyCorrelatedRemovedVariable <- colnames(DPA.Predictors.Numeric)[DPA_HighlyCorrelated[j]]
  print(paste0("Variable ",
               j,
               " for removal: ",
               DPA_HighlyCorrelatedRemovedVariable))
  }

  ##################################
  # Filtering out columns with high correlation
  #################################
  DPA_ExcludedHighCorrelation <- DPA[,-DPA_HighlyCorrelated]

  ##################################
  # Gathering descriptive statistics
  ##################################
  (DPA_ExcludedHighCorrelation_Skimmed <- skim(DPA_ExcludedHighCorrelation))

}

1.3.4 Linear Dependencies

[A] No linear dependencies noted for any subsets of variables using the preprocessing summary from the caret package applying the findLinearCombos method which utilizes the QR decomposition of a matrix to enumerate sets of linear combinations (if they exist).

Code Chunk | Output

##################################
# Loading dataset
##################################
DPA <- Solubility_Train

##################################
# Listing all predictors
##################################
DPA.Predictors <- DPA[,!names(DPA) %in% c("Log_Solubility_Class")]

##################################
# Listing all numeric predictors
##################################
DPA.Predictors.Numeric <- DPA.Predictors[,sapply(DPA.Predictors, is.numeric)]

##################################
# Identifying the linearly dependent variables
##################################
DPA_LinearlyDependent <- findLinearCombos(DPA.Predictors.Numeric)

(DPA_LinearlyDependentCount <- length(DPA_LinearlyDependent$linearCombos))

## [1] 0

if (DPA_LinearlyDependentCount == 0) {
  print("No linearly dependent predictors noted.")
} else {
  print(paste0("Linear dependency observed for ",
               (DPA_LinearlyDependentCount),
               " subset(s) of numeric variable(s)."))

  for (i in 1:DPA_LinearlyDependentCount) {
    DPA_LinearlyDependentSubset <- colnames(DPA.Predictors.Numeric)[DPA_LinearlyDependent$linearCombos[[i]]]
    print(paste0("Linear dependent variable(s) for subset ",
                 i,
                 " include: ",
                 DPA_LinearlyDependentSubset))
  }

}

## [1] "No linearly dependent predictors noted."

##################################
# Identifying the linearly dependent variables for removal
##################################

if (DPA_LinearlyDependentCount > 0) {
  DPA_LinearlyDependent <- findLinearCombos(DPA.Predictors.Numeric)

  DPA_LinearlyDependentForRemoval <- length(DPA_LinearlyDependent$remove)

  print(paste0("Linear dependency can be resolved by removing ",
               (DPA_LinearlyDependentForRemoval),
               " numeric variable(s)."))

  for (j in 1:DPA_LinearlyDependentForRemoval) {
  DPA_LinearlyDependentRemovedVariable <- colnames(DPA.Predictors.Numeric)[DPA_LinearlyDependent$remove[j]]
  print(paste0("Variable ",
               j,
               " for removal: ",
               DPA_LinearlyDependentRemovedVariable))
  }

  ##################################
  # Filtering out columns with linear dependency
  #################################
  DPA_ExcludedLinearlyDependent <- DPA[,-DPA_LinearlyDependent$remove]

  ##################################
  # Gathering descriptive statistics
  ##################################
  (DPA_ExcludedLinearlyDependent_Skimmed <- skim(DPA_ExcludedLinearlyDependent))

}

1.3.5 Pre-Processed Dataset

[A] 1267 rows (observations)
     [A.1] Train Set = 951 observations
     [A.2] Test Set = 316 observations

[B] 5 columns (variables)
     [B.1] 1/5 response = Log_Solubility_Class variable (factor)
            [B.1.1] Levels = Log_Solubility_Class=Low < Log_Solubility_Class=High
     [B.2] 4/5 predictors = All remaining variables (4/4 numeric)

[C] Pre-processing actions applied:
     [C.1] No predictors removed due to zero or near-zero variance, high correlation or linear dependencies

Code Chunk | Output

##################################
# Creating the pre-modelling
# train set
##################################
PMA_PreModelling_Train <- Solubility_Train

##################################
# Gathering descriptive statistics
##################################
(PMA_PreModelling_Train_Skimmed <- skim(PMA_PreModelling_Train))

Data summary

Name	PMA_PreModelling_Train
Number of rows	951
Number of columns	5
_______________________
Column type frequency:
factor	1
numeric	4
________________________
Group variables	None

Variable type: factor

skim_variable	n_missing	complete_rate	ordered	n_unique	top_counts
Log_Solubility_Class	0	1	FALSE	2	Hig: 524, Low: 427

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
HydrophilicFactor	0	1	-0.02	1.13	-0.98	-0.76	-0.31	0.31	13.48	▇▁▁▁▁
NumAtoms	0	1	25.51	12.61	5.00	17.00	22.00	31.00	94.00	▇▆▂▁▁
NumNonHAtoms	0	1	13.16	6.50	2.00	8.00	12.00	17.00	47.00	▇▆▂▁▁
NumCarbon	0	1	9.89	5.29	1.00	6.00	9.00	12.00	33.00	▇▇▃▁▁

###################################
# Verifying the data dimensions
# for the train set
###################################
dim(PMA_PreModelling_Train)

## [1] 951   5

##################################
# Creating the pre-modelling
# test set
##################################
PMA_PreModelling_Test <- Solubility_Test

##################################
# Gathering descriptive statistics
##################################
(PMA_PreModelling_Test_Skimmed <- skim(PMA_PreModelling_Test))

Data summary

Name	PMA_PreModelling_Test
Number of rows	316
Number of columns	5
_______________________
Column type frequency:
factor	1
numeric	4
________________________
Group variables	None

Variable type: factor

skim_variable	n_missing	complete_rate	ordered	n_unique	top_counts
Log_Solubility_Class	0	1	FALSE	2	Hig: 173, Low: 143

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
HydrophilicFactor	0	1	-0.10	0.98	-0.99	-0.77	-0.4	0.21	5	▇▂▁▁▁
NumAtoms	0	1	24.60	11.89	5.00	17.00	22.0	29.00	68	▅▇▂▁▁
NumNonHAtoms	0	1	12.71	6.01	3.00	8.00	11.0	16.00	33	▇▇▅▂▁
NumCarbon	0	1	9.78	5.08	1.00	6.00	8.0	12.00	24	▃▇▃▂▁

###################################
# Verifying the data dimensions
# for the test set
###################################
dim(PMA_PreModelling_Test)

## [1] 316   5

1.4 Data Exploration

[A] All numeric variables demonstrated differential relationships with the Log_Solubility_Class response variable:
     [A.1] NumNonHAtoms variable (numeric)
     [A.2] NumCarbon variable (numeric)
     [A.3] HydrophilicFactor variable (numeric)
     [A.4] NumAtoms variable (numeric)

Code Chunk | Output

##################################
# Loading dataset
##################################
EDA <- PMA_PreModelling_Train

##################################
# Listing all predictors
##################################
EDA.Predictors <- EDA[,!names(EDA) %in% c("Log_Solubility_Class")]

##################################
# Listing all numeric predictors
##################################
EDA.Predictors.Numeric <- EDA.Predictors[,sapply(EDA.Predictors, is.numeric)]
ncol(EDA.Predictors.Numeric)

## [1] 4

names(EDA.Predictors.Numeric)

## [1] "HydrophilicFactor" "NumAtoms"          "NumNonHAtoms"     
## [4] "NumCarbon"

##################################
# Formulating the box plots
##################################
featurePlot(x = EDA.Predictors.Numeric,
            y = EDA$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|")

1.5 Predictive Model Development

1.5.1 Logistic Regression Without Skewness and Outlier Treatment (LR_REF)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

[A] The logistic regression model from the stats package was implemented through the caret package without any treatments applied for both data skewness and outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.87475

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.88447

Code Chunk | Output

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

##################################
# Treating data skewness
# for the train set
##################################
# No actions applied

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)],
            y = PMA_PreModelling_Train_LR$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_REF Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)], 
            y = PMA_PreModelling_Train_LR$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_REF Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_Tune <- train(x = PMA_PreModelling_Train_LR[,!names(PMA_PreModelling_Train_LR) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8747542  0.7001107  0.8397678

LR_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##           4.04521            1.16804            0.05341           -0.28105  
##         NumCarbon  
##          -0.14277  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 834.6     AIC: 844.6

LR_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD    SpecSD
## 1      none 0.8747542 0.7001107 0.8397678 0.03649375 0.06155467 0.0638174

(LR_Train_ROCCurveAUC <- LR_Tune$results$ROC)

## [1] 0.8747542

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_VarImp <- varImp(LR_Tune, scale = TRUE)
plot(LR_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
# No actions applied

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
##################################
# Formulating the box plots
##################################
featurePlot(x = PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)],
            y = PMA_PreModelling_Test_LR$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_REF Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)], 
            y = PMA_PreModelling_Test_LR$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_REF Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_Test <- data.frame(LR_Observed = PMA_PreModelling_Test_LR$Log_Solubility_Class,
                      LR_Predicted = predict(LR_Tune,
                      PMA_PreModelling_Test_LR[,!names(PMA_PreModelling_Test_LR) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_Test

##      LR_Observed LR_Predicted.Low LR_Predicted.High
## 20          High     0.0336865657       0.966313434
## 21          High     0.0153183414       0.984681659
## 23          High     0.0643972187       0.935602781
## 25          High     0.0285205288       0.971479471
## 28          High     0.0798066290       0.920193371
## 31          High     0.0718016713       0.928198329
## 32          High     0.1028972691       0.897102731
## 33          High     0.1274812074       0.872518793
## 34          High     0.1274812074       0.872518793
## 37          High     0.3233789907       0.676621009
## 38          High     0.3233789907       0.676621009
## 42          High     0.5594454651       0.440554535
## 49          High     0.2002171263       0.799782874
## 54          High     0.0684240570       0.931575943
## 55          High     0.0181095804       0.981890420
## 58          High     0.3886335524       0.611366448
## 60          High     0.1402991233       0.859700877
## 61          High     0.1274812074       0.872518793
## 65          High     0.0864344530       0.913565547
## 69          High     0.4637906536       0.536209346
## 73          High     0.0029275515       0.997072449
## 86          High     0.1127673204       0.887232680
## 90          High     0.0962059194       0.903794081
## 91          High     0.0047906462       0.995209354
## 93          High     0.0962059194       0.903794081
## 96          High     0.0047906462       0.995209354
## 98          High     0.1059035446       0.894096455
## 100         High     0.1131780647       0.886821935
## 104         High     0.9327897176       0.067210282
## 112         High     0.2328162785       0.767183721
## 115         High     0.8387113746       0.161288625
## 119         High     0.1836571689       0.816342831
## 128         High     0.1836571689       0.816342831
## 130         High     0.0190842020       0.980915798
## 139         High     0.0190842020       0.980915798
## 143         High     0.0563130978       0.943686902
## 145         High     0.1531943560       0.846805644
## 146         High     0.1836571689       0.816342831
## 149         High     0.1969074264       0.803092574
## 150         High     0.1674558482       0.832544152
## 152         High     0.1274812074       0.872518793
## 157         High     0.4426510503       0.557348950
## 161         High     0.2961059613       0.703894039
## 162         High     0.0061418555       0.993858144
## 166         High     0.4589405370       0.541059463
## 167         High     0.1909996198       0.809000380
## 173         High     0.1530869135       0.846913086
## 176         High     0.1836571689       0.816342831
## 182         High     0.0613936617       0.938606338
## 187         High     0.0441054546       0.955894545
## 190         High     0.0121865069       0.987813493
## 194         High     0.0576402753       0.942359725
## 195         High     0.2423869901       0.757613010
## 201         High     0.1307021191       0.869297881
## 207         High     0.1726530538       0.827346946
## 208         High     0.4277335519       0.572266448
## 215         High     0.0718016713       0.928198329
## 222         High     0.3047752395       0.695224760
## 224         High     0.2013625086       0.798637491
## 231         High     0.6523674422       0.347632558
## 236         High     0.1398934495       0.860106551
## 237         High     0.0890185417       0.910981458
## 240         High     0.2046581225       0.795341878
## 243         High     0.1307021191       0.869297881
## 248         High     0.2423869901       0.757613010
## 251         High     0.8035890142       0.196410986
## 256         High     0.4398434030       0.560156597
## 258         High     0.2233967470       0.776603253
## 262         High     0.4277335519       0.572266448
## 266         High     0.4628428824       0.537157118
## 272         High     0.4989075140       0.501092486
## 280         High     0.3289791156       0.671020884
## 283         High     0.3828696538       0.617130346
## 286         High     0.4495928586       0.550407141
## 287         High     0.2327138907       0.767286109
## 289         High     0.1292348977       0.870765102
## 290         High     0.3529766973       0.647023303
## 298         High     0.2935585665       0.706441434
## 305         High     0.3346274939       0.665372506
## 306         High     0.1894718660       0.810528134
## 312         High     0.1237091711       0.876290829
## 320         High     0.2893537234       0.710646277
## 325         High     0.2038752816       0.796124718
## 332         High     0.0802850496       0.919714950
## 333         High     0.3890415450       0.610958455
## 335         High     0.2935585665       0.706441434
## 339         High     0.6696503319       0.330349668
## 346         High     0.3986640566       0.601335943
## 347         High     0.0860863900       0.913913610
## 350         High     0.3277438405       0.672256160
## 353         High     0.3910406088       0.608959391
## 358         High     0.3733526855       0.626647315
## 365         High     0.2662005002       0.733799500
## 367         High     0.2182970608       0.781702939
## 370         High     0.0287403751       0.971259625
## 379         High     0.1042586676       0.895741332
## 386         High     0.3129499183       0.687050082
## 394         High     0.5395683832       0.460431617
## 396         High     0.1610214870       0.838978513
## 400         High     0.0890185417       0.910981458
## 404         High     0.0721585562       0.927841444
## 405         High     0.5330445369       0.466955463
## 413         High     0.1852494698       0.814750530
## 415         High     0.4082548314       0.591745169
## 417         High     0.2173598482       0.782640152
## 418         High     0.4530410351       0.546958965
## 423         High     0.2838563728       0.716143627
## 434         High     0.3397469664       0.660253034
## 437         High     0.2188160386       0.781183961
## 440         High     0.3550137284       0.644986272
## 449         High     0.4032385216       0.596761478
## 450         High     0.3044787370       0.695521263
## 457         High     0.4032385216       0.596761478
## 467         High     0.2827750579       0.717224942
## 469         High     0.0818798441       0.918120156
## 474         High     0.9175649757       0.082435024
## 475         High     0.8729656154       0.127034385
## 485         High     0.1139410564       0.886058944
## 504          Low     0.1708923019       0.829107698
## 511          Low     0.6509099069       0.349090093
## 512          Low     0.4402430307       0.559756969
## 517          Low     0.0275983874       0.972401613
## 519          Low     0.7833906241       0.216609376
## 520          Low     0.1216573714       0.878342629
## 522          Low     0.9204361063       0.079563894
## 527          Low     0.6497262336       0.350273766
## 528          Low     0.2689383818       0.731061618
## 529          Low     0.3420093611       0.657990639
## 537          Low     0.1601409934       0.839859007
## 540          Low     0.8991542777       0.100845722
## 541          Low     0.4407363999       0.559263600
## 547          Low     0.8997547622       0.100245238
## 550          Low     0.3840800947       0.615919905
## 555          Low     0.5881995873       0.411800413
## 564          Low     0.0792550489       0.920744951
## 570          Low     0.4185241373       0.581475863
## 573          Low     0.2767634485       0.723236551
## 575          Low     0.4407363999       0.559263600
## 578          Low     0.1997361134       0.800263887
## 581          Low     0.2767634485       0.723236551
## 585          Low     0.3394330523       0.660566948
## 590          Low     0.7555077932       0.244492207
## 601          Low     0.9234128549       0.076587145
## 602          Low     0.6369498782       0.363050122
## 607          Low     0.6234820573       0.376517943
## 610          Low     0.6108630426       0.389136957
## 618          Low     0.7753385721       0.224661428
## 624          Low     0.3394330523       0.660566948
## 626          Low     0.2798585686       0.720141431
## 627          Low     0.3204248034       0.679575197
## 634          Low     0.5983101185       0.401689881
## 640          Low     0.9898595316       0.010140468
## 642          Low     0.1832068271       0.816793173
## 643          Low     0.5081035274       0.491896473
## 644          Low     0.8116432284       0.188356772
## 645          Low     0.7002023850       0.299797615
## 646          Low     0.6507453243       0.349254676
## 647          Low     0.8301016048       0.169898395
## 652          Low     0.2054268163       0.794573184
## 658          Low     0.6607108311       0.339289169
## 659          Low     0.7510505757       0.248949424
## 660          Low     0.9161958942       0.083804106
## 664          Low     0.3408694824       0.659130518
## 666          Low     0.4192476727       0.580752327
## 667          Low     0.8681760956       0.131823904
## 675          Low     0.5081035274       0.491896473
## 680          Low     0.9926469280       0.007353072
## 681          Low     0.9166739748       0.083326025
## 687          Low     0.7797970867       0.220202913
## 694          Low     0.7788178093       0.221182191
## 697          Low     0.4799232251       0.520076775
## 701          Low     0.3033114910       0.696688509
## 705          Low     0.9802413586       0.019758641
## 707          Low     0.6893653986       0.310634601
## 710          Low     0.6311139386       0.368886061
## 716          Low     0.9129372667       0.087062733
## 719          Low     0.9266307028       0.073369297
## 720          Low     0.9743859647       0.025614035
## 725          Low     0.9786074660       0.021392534
## 727          Low     0.3033114910       0.696688509
## 730          Low     0.4471810591       0.552818941
## 738          Low     0.7769124419       0.223087558
## 745          Low     0.5748922115       0.425107789
## 748          Low     0.6952562425       0.304743758
## 751          Low     0.9647413723       0.035258628
## 756          Low     0.7139313964       0.286068604
## 766          Low     0.8661335675       0.133866432
## 769          Low     0.5137458692       0.486254131
## 783          Low     0.8026569527       0.197343047
## 785          Low     0.8373166692       0.162683331
## 790          Low     0.9240417860       0.075958214
## 793          Low     0.8026569527       0.197343047
## 795          Low     0.9935450908       0.006454909
## 796          Low     0.9835769734       0.016423027
## 797          Low     0.6437243296       0.356275670
## 801          Low     0.6958616855       0.304138315
## 811          Low     0.4691432530       0.530856747
## 812          Low     0.9617475976       0.038252402
## 815          Low     0.9376097372       0.062390263
## 816          Low     0.7074523261       0.292547674
## 817          Low     0.9234880845       0.076511916
## 824          Low     0.8349276989       0.165072301
## 825          Low     0.8349276989       0.165072301
## 826          Low     0.8349276989       0.165072301
## 830          Low     0.8513492063       0.148650794
## 837          Low     0.9069592513       0.093040749
## 838          Low     0.7074523261       0.292547674
## 844          Low     0.8819618824       0.118038118
## 845          Low     0.9748569182       0.025143082
## 847          Low     0.9174469864       0.082553014
## 850          Low     0.8637880842       0.136211916
## 852          Low     0.8888561225       0.111143878
## 853          Low     0.8888561225       0.111143878
## 861          Low     0.9101752188       0.089824781
## 868          Low     0.9795950024       0.020404998
## 874          Low     0.9281279972       0.071872003
## 879         High     0.1529569733       0.847043027
## 895         High     0.0463424895       0.953657511
## 899         High     0.0001178774       0.999882123
## 903         High     0.0463424895       0.953657511
## 917         High     0.0423986925       0.957601308
## 927         High     0.0684240570       0.931575943
## 929         High     0.1530869135       0.846913086
## 931         High     0.0684240570       0.931575943
## 933         High     0.4172979708       0.582702029
## 944         High     0.0720404149       0.927959585
## 947         High     0.0962059194       0.903794081
## 949         High     0.2223733535       0.777626646
## 953         High     0.0273184034       0.972681597
## 958         High     0.4446072150       0.555392785
## 961         High     0.0326363542       0.967363646
## 963         High     0.1566911276       0.843308872
## 964         High     0.1307021191       0.869297881
## 973         High     0.0688034909       0.931196509
## 976         High     0.0792550489       0.920744951
## 977         High     0.2423869901       0.757613010
## 980         High     0.2318664434       0.768133557
## 983         High     0.5516380511       0.448361949
## 984         High     0.2423869901       0.757613010
## 986         High     0.1237091711       0.876290829
## 989         High     0.2142824472       0.785717553
## 991         High     0.0363618409       0.963638159
## 996         High     0.0980702886       0.901929711
## 997         High     0.4019257860       0.598074214
## 999         High     0.0643972187       0.935602781
## 1000        High     0.0622947220       0.937705278
## 1003        High     0.0890185417       0.910981458
## 1008        High     0.1343419103       0.865658090
## 1009        High     0.3828696538       0.617130346
## 1014        High     0.0688164915       0.931183509
## 1015        High     0.5341066328       0.465893367
## 1040        High     0.2006168147       0.799383185
## 1042        High     0.3777393739       0.622260626
## 1043        High     0.6482206500       0.351779350
## 1050        High     0.1245951154       0.875404885
## 1052        High     0.2328000692       0.767199931
## 1056        High     0.0302052801       0.969794720
## 1070        High     0.5790424430       0.420957557
## 1073        High     0.4142959793       0.585704021
## 1074        High     0.1993893842       0.800610616
## 1079        High     0.3579641321       0.642035868
## 1080        High     0.4873669707       0.512633029
## 1085        High     0.1155517838       0.884448216
## 1087        High     0.6888614637       0.311138536
## 1096        High     0.9458132723       0.054186728
## 1099        High     0.5089195538       0.491080446
## 1100        High     0.6024802982       0.397519702
## 1102        High     0.0512009711       0.948799029
## 1107         Low     0.3749354744       0.625064526
## 1109         Low     0.7661888954       0.233811105
## 1114         Low     0.4002856069       0.599714393
## 1118         Low     0.4347382476       0.565261752
## 1123         Low     0.4690012962       0.530998704
## 1132         Low     0.8206401102       0.179359890
## 1134         Low     0.6312389825       0.368761018
## 1137         Low     0.3204248034       0.679575197
## 1154         Low     0.3204248034       0.679575197
## 1155         Low     0.6060432019       0.393956798
## 1157         Low     0.8609402930       0.139059707
## 1162         Low     0.5081035274       0.491896473
## 1164         Low     0.1832068271       0.816793173
## 1171         Low     0.9647413723       0.035258628
## 1172         Low     0.4426510503       0.557348950
## 1175         Low     0.7347521726       0.265247827
## 1177         Low     0.5669973713       0.433002629
## 1179         Low     0.9495401865       0.050459814
## 1183         Low     0.2484602589       0.751539741
## 1185         Low     0.9700991083       0.029900892
## 1189         Low     0.9353149148       0.064685085
## 1211         Low     0.7090081559       0.290991844
## 1218         Low     0.9962485352       0.003751465
## 1224         Low     0.3641947857       0.635805214
## 1225         Low     0.3033114910       0.696688509
## 1227         Low     0.9477186443       0.052281356
## 1232         Low     0.9906240752       0.009375925
## 1235         Low     0.7850592628       0.214940737
## 1238         Low     0.8244394114       0.175560589
## 1240         Low     0.8970123273       0.102987673
## 1241         Low     0.7090081559       0.290991844
## 1248         Low     0.8513492063       0.148650794
## 1258         Low     0.7074523261       0.292547674
## 1261         Low     0.8640693634       0.135930637
## 1263         Low     0.8349276989       0.165072301
## 1269         Low     0.9566595486       0.043340451
## 1270         Low     0.9911281573       0.008871843
## 1271         Low     0.8637880842       0.136211916
## 1272         Low     0.8888561225       0.111143878
## 1280         Low     0.9281279972       0.071872003
## 1286         Low     0.9890962672       0.010903733
## 1287         Low     0.9901901712       0.009809829
## 1289         Low     0.9548963849       0.045103615
## 1290         Low     0.8888561225       0.111143878
## 1291        High     0.2132741120       0.786725888
## 1294        High     0.7243151562       0.275684844
## 1305         Low     0.9635883387       0.036411661
## 1308        High     0.8468798982       0.153120102

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_Test_ROC <- roc(response = LR_Test$LR_Observed,
             predictor = LR_Test$LR_Predicted.High,
             levels = rev(levels(LR_Test$LR_Observed)))

(LR_Test_ROCCurveAUC <- auc(LR_Test_ROC)[1])

## [1] 0.8844739

1.5.2 Logistic Regression With Box-Cox Transformation (LR_BCT)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Box-Cox Transformation applies a family of power transformations such that the modified values are a monotonic function of the observations over some admissible range and indexed by an optimal parameter lambda. The process involves determining the value of lambda which would result to the highest correlation between the Box-Cox-transformed values and the z-scores of the observations’ ordered indices. The original data set is recomputed through the same power transformations given the optimized lambda value. The method is restricted to positive values only.

[A] The logistic regression model from the stats package was implemented through the caret package with Box-Cox transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.88878

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89676

Code Chunk | Output

##################################
# Adding an offset to adjust the
# range of values to only positive values
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  PMA_PreModelling_Train[,i] <- PMA_PreModelling_Train[,i]+1
}

for (i in 1:(ncol(PMA_PreModelling_Test)-1)){
  PMA_PreModelling_Test[,i] <- PMA_PreModelling_Test[,i]+1
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
Transform_BoxCox <- preProcess(PMA_PreModelling_Train_LR, method = c("BoxCox"))
PMA_PreModelling_Train_LR_BCT <- predict(Transform_BoxCox, PMA_PreModelling_Train_LR.Numeric)
PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_BCT[,sapply(PMA_PreModelling_Train_LR_BCT, is.numeric)],
            y = PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_BCT Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_BCT[,sapply(PMA_PreModelling_Train_LR_BCT, is.numeric)], 
            y = PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_BCT Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_BCT_Tune <- train(x = PMA_PreModelling_Train_LR_BCT[,!names(PMA_PreModelling_Train_LR_BCT) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_BCT$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_BCT_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8887838  0.7281285  0.8283382

LR_BCT_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##           10.7676             1.3301             1.4474            -5.4764  
##         NumCarbon  
##           -0.1236  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 779.5     AIC: 789.5

LR_BCT_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8887838 0.7281285 0.8283382 0.03378421 0.07358024 0.06110159

(LR_BCT_Train_ROCCurveAUC <- LR_BCT_Tune$results$ROC)

## [1] 0.8887838

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_BCT_VarImp <- varImp(LR_BCT_Tune, scale = TRUE)
plot(LR_BCT_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_BCT <- predict(Transform_BoxCox, PMA_PreModelling_Test_LR.Numeric)
PMA_PreModelling_Test_LR_BCT$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_BCT[,sapply(PMA_PreModelling_Test_LR_BCT, is.numeric)],
            y = PMA_PreModelling_Test_LR_BCT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_BCT Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_BCT[,sapply(PMA_PreModelling_Test_LR_BCT, is.numeric)], 
            y = PMA_PreModelling_Test_LR_BCT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_BCT Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_BCT$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_BCT_Test <- data.frame(LR_BCT_Observed = PMA_PreModelling_Test_LR_BCT$Log_Solubility_Class,
                      LR_BCT_Predicted = predict(LR_BCT_Tune,
                      PMA_PreModelling_Test_LR_BCT[,!names(PMA_PreModelling_Test_LR_BCT) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_BCT_Test

##      LR_BCT_Observed LR_BCT_Predicted.Low LR_BCT_Predicted.High
## 20              High         0.0106788478           0.989321152
## 21              High         0.0071362537           0.992863746
## 23              High         0.0393459886           0.960654011
## 25              High         0.0121378710           0.987862129
## 28              High         0.0624328061           0.937567194
## 31              High         0.0138721555           0.986127844
## 32              High         0.0247401461           0.975259854
## 33              High         0.0461679379           0.953832062
## 34              High         0.0461679379           0.953832062
## 37              High         0.3571115534           0.642888447
## 38              High         0.3571115534           0.642888447
## 42              High         0.4980215283           0.501978472
## 49              High         0.1438747853           0.856125215
## 54              High         0.0165462619           0.983453738
## 55              High         0.0166769554           0.983323045
## 58              High         0.4550864082           0.544913592
## 60              High         0.0600003953           0.939999605
## 61              High         0.0461679379           0.953832062
## 65              High         0.0447692681           0.955230732
## 69              High         0.5416238123           0.458376188
## 73              High         0.0197460856           0.980253914
## 86              High         0.0422336568           0.957766343
## 90              High         0.0321165091           0.967883491
## 91              High         0.0094919686           0.990508031
## 93              High         0.0321165091           0.967883491
## 96              High         0.0094919686           0.990508031
## 98              High         0.0366929536           0.963307046
## 100             High         0.0542108436           0.945789156
## 104             High         0.8776746097           0.122325390
## 112             High         0.1983421147           0.801657885
## 115             High         0.8831774718           0.116822528
## 119             High         0.1241227426           0.875877257
## 128             High         0.1241227426           0.875877257
## 130             High         0.0181786697           0.981821330
## 139             High         0.0181786697           0.981821330
## 143             High         0.0378887829           0.962111217
## 145             High         0.0730209400           0.926979060
## 146             High         0.1241227426           0.875877257
## 149             High         0.1481173988           0.851882601
## 150             High         0.0964081466           0.903591853
## 152             High         0.0461679379           0.953832062
## 157             High         0.5424164866           0.457583513
## 161             High         0.2169241823           0.783075818
## 162             High         0.0081207937           0.991879206
## 166             High         0.5350267240           0.464973276
## 167             High         0.1238179689           0.876182031
## 173             High         0.0832659662           0.916734034
## 176             High         0.1241227426           0.875877257
## 182             High         0.0051278105           0.994872189
## 187             High         0.0422067189           0.957793281
## 190             High         0.0180259773           0.981974023
## 194             High         0.0157103850           0.984289615
## 195             High         0.1420659840           0.857934016
## 201             High         0.0566739425           0.943326058
## 207             High         0.0923516972           0.907648303
## 208             High         0.5326871755           0.467312825
## 215             High         0.0138721555           0.986127844
## 222             High         0.2009338031           0.799066197
## 224             High         0.2021924279           0.797807572
## 231             High         0.7632973587           0.236702641
## 236             High         0.0948388987           0.905161101
## 237             High         0.0164527338           0.983547266
## 240             High         0.1992633468           0.800736653
## 243             High         0.0566739425           0.943326058
## 248             High         0.1420659840           0.857934016
## 251             High         0.7380831273           0.261916873
## 256             High         0.3579586896           0.642041310
## 258             High         0.1476808290           0.852319171
## 262             High         0.5326871755           0.467312825
## 266             High         0.4699388730           0.530061127
## 272             High         0.5770850405           0.422914959
## 280             High         0.3130800631           0.686919937
## 283             High         0.4439245125           0.556075487
## 286             High         0.4514948386           0.548505161
## 287             High         0.1336742035           0.866325796
## 289             High         0.1193414649           0.880658535
## 290             High         0.3335117710           0.666488229
## 298             High         0.3052569144           0.694743086
## 305             High         0.2800013287           0.719998671
## 306             High         0.2260049236           0.773995076
## 312             High         0.1032698105           0.896730190
## 320             High         0.3076680491           0.692331951
## 325             High         0.1842236116           0.815776388
## 332             High         0.0586131421           0.941386858
## 333             High         0.4769698970           0.523030103
## 335             High         0.3052569144           0.694743086
## 339             High         0.8166001784           0.183399822
## 346             High         0.2935888502           0.706411150
## 347             High         0.0320240534           0.967975947
## 350             High         0.4009958389           0.599004161
## 353             High         0.3131508028           0.686849197
## 358             High         0.2701478279           0.729852172
## 365             High         0.2156848824           0.784315118
## 367             High         0.2112425750           0.788757425
## 370             High         0.0509989361           0.949001064
## 379             High         0.1262417880           0.873758212
## 386             High         0.4443738084           0.555626192
## 394             High         0.5041699492           0.495830051
## 396             High         0.1311385627           0.868861437
## 400             High         0.0164527338           0.983547266
## 404             High         0.0393672403           0.960632760
## 405             High         0.6364127374           0.363587263
## 413             High         0.1147530425           0.885246957
## 415             High         0.3508766249           0.649123375
## 417             High         0.2248338620           0.775166138
## 418             High         0.4108293636           0.589170636
## 423             High         0.2838050184           0.716194982
## 434             High         0.2353807171           0.764619283
## 437             High         0.2990469896           0.700953010
## 440             High         0.3983575610           0.601642439
## 449             High         0.3740547041           0.625945296
## 450             High         0.2800405789           0.719959421
## 457             High         0.3740547041           0.625945296
## 467             High         0.3477870747           0.652212925
## 469             High         0.2411307699           0.758869230
## 474             High         0.9420572222           0.057942778
## 475             High         0.9065732264           0.093426774
## 485             High         0.1233535430           0.876646457
## 504              Low         0.2527906760           0.747209324
## 511              Low         0.6865391505           0.313460849
## 512              Low         0.4047166486           0.595283351
## 517              Low         0.0656989760           0.934301024
## 519              Low         0.6847887241           0.315211276
## 520              Low         0.0419643779           0.958035622
## 522              Low         0.9512569299           0.048743070
## 527              Low         0.5950600444           0.404939956
## 528              Low         0.3666971317           0.633302868
## 529              Low         0.3033258986           0.696674101
## 537              Low         0.0894377504           0.910562250
## 540              Low         0.9299392609           0.070060739
## 541              Low         0.6976605213           0.302339479
## 547              Low         0.9224917546           0.077508245
## 550              Low         0.6120118529           0.387988147
## 555              Low         0.4797523411           0.520247659
## 564              Low         0.0198374990           0.980162501
## 570              Low         0.3404075983           0.659592402
## 573              Low         0.2308988075           0.769101192
## 575              Low         0.6976605213           0.302339479
## 578              Low         0.1980215186           0.801978481
## 581              Low         0.2308988075           0.769101192
## 585              Low         0.3361800443           0.663819956
## 590              Low         0.7115152158           0.288484784
## 601              Low         0.7926652156           0.207334784
## 602              Low         0.6243017798           0.375698220
## 607              Low         0.5287426671           0.471257333
## 610              Low         0.5139990677           0.486000932
## 618              Low         0.7102111890           0.289788811
## 624              Low         0.3361800443           0.663819956
## 626              Low         0.2006616946           0.799338305
## 627              Low         0.3402550529           0.659744947
## 634              Low         0.6211336105           0.378866390
## 640              Low         0.9838055034           0.016194497
## 642              Low         0.1760170355           0.823982965
## 643              Low         0.7851585980           0.214841402
## 644              Low         0.8721155291           0.127884471
## 645              Low         0.7767636349           0.223236365
## 646              Low         0.8881478956           0.111852104
## 647              Low         0.7286976220           0.271302378
## 652              Low         0.1652311255           0.834768874
## 658              Low         0.5649909662           0.435009034
## 659              Low         0.9358548026           0.064145197
## 660              Low         0.8494685582           0.150531442
## 664              Low         0.3462913416           0.653708658
## 666              Low         0.4573944495           0.542605550
## 667              Low         0.8366650170           0.163334983
## 675              Low         0.7851585980           0.214841402
## 680              Low         0.9649799678           0.035020032
## 681              Low         0.8599408367           0.140059163
## 687              Low         0.9460336996           0.053966300
## 694              Low         0.7144664403           0.285533560
## 697              Low         0.5847904454           0.415209555
## 701              Low         0.4828825333           0.517117467
## 705              Low         0.9090396249           0.090960375
## 707              Low         0.6414060837           0.358593916
## 710              Low         0.7840992120           0.215900788
## 716              Low         0.8522709843           0.147729016
## 719              Low         0.9417254613           0.058274539
## 720              Low         0.9812037744           0.018796226
## 725              Low         0.9797183384           0.020281662
## 727              Low         0.4828825333           0.517117467
## 730              Low         0.5419442378           0.458055762
## 738              Low         0.8926103955           0.107389604
## 745              Low         0.8485016772           0.151498323
## 748              Low         0.8291228102           0.170877190
## 751              Low         0.9156090270           0.084390973
## 756              Low         0.6729571755           0.327042824
## 766              Low         0.7161626263           0.283837374
## 769              Low         0.6479837562           0.352016244
## 783              Low         0.8822621938           0.117737806
## 785              Low         0.9637029474           0.036297053
## 790              Low         0.8580730609           0.141926939
## 793              Low         0.8822621938           0.117737806
## 795              Low         0.9888028122           0.011197188
## 796              Low         0.9830892650           0.016910735
## 797              Low         0.5255268720           0.474473128
## 801              Low         0.7934333977           0.206566602
## 811              Low         0.5931980901           0.406801910
## 812              Low         0.9094790547           0.090520945
## 815              Low         0.9737933608           0.026206639
## 816              Low         0.7873874858           0.212612514
## 817              Low         0.9838411522           0.016158848
## 824              Low         0.8906357926           0.109364207
## 825              Low         0.8906357926           0.109364207
## 826              Low         0.8906357926           0.109364207
## 830              Low         0.9674044746           0.032595525
## 837              Low         0.9803346321           0.019665368
## 838              Low         0.7873874858           0.212612514
## 844              Low         0.9746818677           0.025318132
## 845              Low         0.9935410672           0.006458933
## 847              Low         0.9348514760           0.065148524
## 850              Low         0.9033168636           0.096683136
## 852              Low         0.9165077717           0.083492228
## 853              Low         0.9165077717           0.083492228
## 861              Low         0.9287872244           0.071212776
## 868              Low         0.9945213964           0.005478604
## 874              Low         0.9399255679           0.060074432
## 879             High         0.0820381758           0.917961824
## 895             High         0.0075429914           0.992457009
## 899             High         0.0003688008           0.999631199
## 903             High         0.0075429914           0.992457009
## 917             High         0.0512730090           0.948726991
## 927             High         0.0165462619           0.983453738
## 929             High         0.0832659662           0.916734034
## 931             High         0.0165462619           0.983453738
## 933             High         0.3637050472           0.636294953
## 944             High         0.0345478479           0.965452152
## 947             High         0.0321165091           0.967883491
## 949             High         0.1830929077           0.816907092
## 953             High         0.0247953563           0.975204644
## 958             High         0.5752647020           0.424735298
## 961             High         0.0114054875           0.988594513
## 963             High         0.1003959629           0.899604037
## 964             High         0.0566739425           0.943326058
## 973             High         0.0493627059           0.950637294
## 976             High         0.0198374990           0.980162501
## 977             High         0.1420659840           0.857934016
## 980             High         0.2975453104           0.702454690
## 983             High         0.6552675251           0.344732475
## 984             High         0.1420659840           0.857934016
## 986             High         0.1032698105           0.896730190
## 989             High         0.1752033347           0.824796665
## 991             High         0.0216816050           0.978318395
## 996             High         0.0213154376           0.978684562
## 997             High         0.4645359623           0.535464038
## 999             High         0.0393459886           0.960654011
## 1000            High         0.0558764682           0.944123532
## 1003            High         0.0164527338           0.983547266
## 1008            High         0.0923518450           0.907648155
## 1009            High         0.4439245125           0.556075487
## 1014            High         0.0345635445           0.965436455
## 1015            High         0.4759596868           0.524040313
## 1040            High         0.1881135515           0.811886448
## 1042            High         0.3560654524           0.643934548
## 1043            High         0.7350851706           0.264914829
## 1050            High         0.1128103317           0.887189668
## 1052            High         0.1590707746           0.840929225
## 1056            High         0.2601441055           0.739855895
## 1070            High         0.5365314855           0.463468514
## 1073            High         0.5067464368           0.493253563
## 1074            High         0.1359320993           0.864067901
## 1079            High         0.4166247667           0.583375233
## 1080            High         0.6521706797           0.347829320
## 1085            High         0.0914183321           0.908581668
## 1087            High         0.5954058977           0.404594102
## 1096            High         0.8884323804           0.111567620
## 1099            High         0.4356482012           0.564351799
## 1100            High         0.7147323089           0.285267691
## 1102            High         0.0756794028           0.924320597
## 1107             Low         0.5814501628           0.418549837
## 1109             Low         0.6608220214           0.339177979
## 1114             Low         0.6200404052           0.379959595
## 1118             Low         0.3315143332           0.668485667
## 1123             Low         0.4710123894           0.528987611
## 1132             Low         0.9166894611           0.083310539
## 1134             Low         0.5681675652           0.431832435
## 1137             Low         0.3402550529           0.659744947
## 1154             Low         0.3402550529           0.659744947
## 1155             Low         0.5465223806           0.453477619
## 1157             Low         0.7298873873           0.270112613
## 1162             Low         0.7851585980           0.214841402
## 1164             Low         0.1760170355           0.823982965
## 1171             Low         0.9156090270           0.084390973
## 1172             Low         0.5424164866           0.457583513
## 1175             Low         0.6006201592           0.399379841
## 1177             Low         0.7100702299           0.289929770
## 1179             Low         0.9122765249           0.087723475
## 1183             Low         0.3352874269           0.664712573
## 1185             Low         0.9229073005           0.077092699
## 1189             Low         0.7983456906           0.201654309
## 1211             Low         0.9189338153           0.081066185
## 1218             Low         0.9896125086           0.010387491
## 1224             Low         0.4190665046           0.580933495
## 1225             Low         0.4828825333           0.517117467
## 1227             Low         0.8952185285           0.104781471
## 1232             Low         0.9862080090           0.013791991
## 1235             Low         0.7011319853           0.298868015
## 1238             Low         0.6722498554           0.327750145
## 1240             Low         0.8403084604           0.159691540
## 1241             Low         0.9189338153           0.081066185
## 1248             Low         0.9674044746           0.032595525
## 1258             Low         0.7873874858           0.212612514
## 1261             Low         0.8966286234           0.103371377
## 1263             Low         0.8906357926           0.109364207
## 1269             Low         0.9567248742           0.043275126
## 1270             Low         0.9871950642           0.012804936
## 1271             Low         0.9033168636           0.096683136
## 1272             Low         0.9165077717           0.083492228
## 1280             Low         0.9399255679           0.060074432
## 1286             Low         0.9964564442           0.003543556
## 1287             Low         0.9967294062           0.003270594
## 1289             Low         0.9574638600           0.042536140
## 1290             Low         0.9165077717           0.083492228
## 1291            High         0.1740583738           0.825941626
## 1294            High         0.8061455726           0.193854427
## 1305             Low         0.8709712683           0.129028732
## 1308            High         0.7071099477           0.292890052

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_BCT_Test_ROC <- roc(response = LR_BCT_Test$LR_BCT_Observed,
             predictor = LR_BCT_Test$LR_BCT_Predicted.High,
             levels = rev(levels(LR_BCT_Test$LR_BCT_Observed)))

(LR_BCT_Test_ROCCurveAUC <- auc(LR_BCT_Test_ROC)[1])

## [1] 0.8967622

1.5.3 Logistic Regression With Yeo-Johnson Transformation (LR_YJT)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Yeo-Johnson Transformation applies a new family of distributions that can be used without restrictions, extending many of the good properties of the Box-Cox power family. Similar to the Box-Cox transformation, the method also estimates the optimal value of lambda but has the ability to transform both positive and negative values by inflating low variance data and deflating high variance data to create a more uniform data set. While there are no restrictions in terms of the applicable values, the interpretability of the transformed values is more diminished as compared to the other methods.

[A] The logistic regression model from the stats package was implemented through the caret package with Yeo-Johnson transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.88070

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89061

Code Chunk | Output

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
Transform_YeoJohnson <- preProcess(PMA_PreModelling_Train_LR, method = c("YeoJohnson"))
PMA_PreModelling_Train_LR_YJT <- predict(Transform_YeoJohnson, PMA_PreModelling_Train_LR.Numeric)
PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_YJT[,sapply(PMA_PreModelling_Train_LR_YJT, is.numeric)],
            y = PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_YJT Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_YJT[,sapply(PMA_PreModelling_Train_LR_YJT, is.numeric)], 
            y = PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_YJT Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_YJT_Tune <- train(x = PMA_PreModelling_Train_LR_YJT[,!names(PMA_PreModelling_Train_LR_YJT) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_YJT$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_YJT_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8807066  0.7187154  0.8283019

LR_YJT_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##            9.3048             1.4325             1.2632            -4.0866  
##         NumCarbon  
##           -0.4926  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 806.5     AIC: 816.5

LR_YJT_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8807066 0.7187154 0.8283019 0.03522898 0.07917814 0.05907546

(LR_YJT_Train_ROCCurveAUC <- LR_YJT_Tune$results$ROC)

## [1] 0.8807066

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_YJT_VarImp <- varImp(LR_YJT_Tune, scale = TRUE)
plot(LR_YJT_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_YJT <- predict(Transform_YeoJohnson, PMA_PreModelling_Test_LR.Numeric)
PMA_PreModelling_Test_LR_YJT$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_YJT[,sapply(PMA_PreModelling_Test_LR_YJT, is.numeric)],
            y = PMA_PreModelling_Test_LR_YJT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_YJT Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_YJT[,sapply(PMA_PreModelling_Test_LR_YJT, is.numeric)], 
            y = PMA_PreModelling_Test_LR_YJT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_YJT Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_YJT$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_YJT_Test <- data.frame(LR_YJT_Observed = PMA_PreModelling_Test_LR_YJT$Log_Solubility_Class,
                      LR_YJT_Predicted = predict(LR_YJT_Tune,
                      PMA_PreModelling_Test_LR_YJT[,!names(PMA_PreModelling_Test_LR_YJT) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_YJT_Test

##      LR_YJT_Observed LR_YJT_Predicted.Low LR_YJT_Predicted.High
## 20              High         0.0111852876            0.98881471
## 21              High         0.0086628052            0.99133719
## 23              High         0.0465501013            0.95344990
## 25              High         0.0150865233            0.98491348
## 28              High         0.0673498546            0.93265015
## 31              High         0.0187893105            0.98121069
## 32              High         0.0385145943            0.96148541
## 33              High         0.0651287215            0.93487128
## 34              High         0.0651287215            0.93487128
## 37              High         0.3795518012            0.62044820
## 38              High         0.3795518012            0.62044820
## 42              High         0.5083619918            0.49163801
## 49              High         0.1725513385            0.82744866
## 54              High         0.0211968888            0.97880311
## 55              High         0.0243796249            0.97562038
## 58              High         0.4639398237            0.53606018
## 60              High         0.0740667556            0.92593324
## 61              High         0.0651287215            0.93487128
## 65              High         0.0454691543            0.95453085
## 69              High         0.5509798165            0.44902018
## 73              High         0.0335583314            0.96644167
## 86              High         0.0553639019            0.94463610
## 90              High         0.0398746182            0.96012538
## 91              High         0.0177295306            0.98227047
## 93              High         0.0398746182            0.96012538
## 96              High         0.0177295306            0.98227047
## 98              High         0.0455680065            0.95443199
## 100             High         0.0607090959            0.93929090
## 104             High         0.8782141160            0.12178588
## 112             High         0.2280878343            0.77191217
## 115             High         0.8818973943            0.11810261
## 119             High         0.1495433520            0.85045665
## 128             High         0.1495433520            0.85045665
## 130             High         0.0265537930            0.97344621
## 139             High         0.0265537930            0.97344621
## 143             High         0.0408285223            0.95917148
## 145             High         0.1010816593            0.89891834
## 146             High         0.1495433520            0.85045665
## 149             High         0.1495628278            0.85043717
## 150             High         0.1223809375            0.87761906
## 152             High         0.0651287215            0.93487128
## 157             High         0.5324789498            0.46752105
## 161             High         0.2353887548            0.76461125
## 162             High         0.0108126332            0.98918737
## 166             High         0.5421256704            0.45787433
## 167             High         0.1557431213            0.84425688
## 173             High         0.1060669672            0.89393303
## 176             High         0.1495433520            0.85045665
## 182             High         0.0089571356            0.99104286
## 187             High         0.0559846274            0.94401537
## 190             High         0.0229926784            0.97700732
## 194             High         0.0164604888            0.98353951
## 195             High         0.1682281293            0.83177187
## 201             High         0.0687190029            0.93128100
## 207             High         0.1097517332            0.89024827
## 208             High         0.5386162406            0.46138376
## 215             High         0.0187893105            0.98121069
## 222             High         0.2334165404            0.76658346
## 224             High         0.1887353848            0.81126462
## 231             High         0.7667001343            0.23329987
## 236             High         0.0956712807            0.90432872
## 237             High         0.0259764904            0.97402351
## 240             High         0.1856240164            0.81437598
## 243             High         0.0687190029            0.93128100
## 248             High         0.1682281293            0.83177187
## 251             High         0.7507400098            0.24925999
## 256             High         0.3828163835            0.61718362
## 258             High         0.1615928275            0.83840717
## 262             High         0.5386162406            0.46138376
## 266             High         0.4414142561            0.55858574
## 272             High         0.5848712015            0.41512880
## 280             High         0.2943400032            0.70566000
## 283             High         0.4656557701            0.53434423
## 286             High         0.4230688180            0.57693118
## 287             High         0.1584876212            0.84151238
## 289             High         0.1001632377            0.89983676
## 290             High         0.3147070899            0.68529291
## 298             High         0.3331168358            0.66688316
## 305             High         0.2845465573            0.71545344
## 306             High         0.2377354619            0.76226454
## 312             High         0.1089806628            0.89101934
## 320             High         0.3282503800            0.67174962
## 325             High         0.1789329323            0.82106707
## 332             High         0.0616433564            0.93835664
## 333             High         0.4825359090            0.51746409
## 335             High         0.3331168358            0.66688316
## 339             High         0.7915253506            0.20847465
## 346             High         0.3337862922            0.66621371
## 347             High         0.0354544470            0.96454555
## 350             High         0.4051825670            0.59481743
## 353             High         0.3366565097            0.66334349
## 358             High         0.3080651712            0.69193483
## 365             High         0.2184421650            0.78155784
## 367             High         0.2078586791            0.79214132
## 370             High         0.0572003924            0.94279961
## 379             High         0.1413219999            0.85867800
## 386             High         0.3773059137            0.62269409
## 394             High         0.4977002808            0.50229972
## 396             High         0.1361516876            0.86384831
## 400             High         0.0259764904            0.97402351
## 404             High         0.0341686706            0.96583133
## 405             High         0.6483833587            0.35161664
## 413             High         0.1298373285            0.87016267
## 415             High         0.3574576992            0.64254230
## 417             High         0.2052920777            0.79470792
## 418             High         0.4180844835            0.58191552
## 423             High         0.3207522384            0.67924776
## 434             High         0.2720066738            0.72799333
## 437             High         0.2932273518            0.70677265
## 440             High         0.3919768236            0.60802318
## 449             High         0.3535353096            0.64646469
## 450             High         0.2675541417            0.73244586
## 457             High         0.3535353096            0.64646469
## 467             High         0.3416031366            0.65839686
## 469             High         0.2412064259            0.75879357
## 474             High         0.9372709045            0.06272910
## 475             High         0.9045788808            0.09542112
## 485             High         0.1400180092            0.85998199
## 504              Low         0.3014552746            0.69854473
## 511              Low         0.7986676581            0.20133234
## 512              Low         0.3948293950            0.60517061
## 517              Low         0.0915588732            0.90844113
## 519              Low         0.6710622179            0.32893778
## 520              Low         0.0593466688            0.94065333
## 522              Low         0.9427595075            0.05724049
## 527              Low         0.6066246995            0.39337530
## 528              Low         0.3047234793            0.69527652
## 529              Low         0.2940689781            0.70593102
## 537              Low         0.1137543356            0.88624566
## 540              Low         0.9257268536            0.07427315
## 541              Low         0.5764299193            0.42357008
## 547              Low         0.9213479392            0.07865206
## 550              Low         0.7714659594            0.22853404
## 555              Low         0.4823762799            0.51762372
## 564              Low         0.0264753239            0.97352468
## 570              Low         0.3415221433            0.65847786
## 573              Low         0.2337758471            0.76622415
## 575              Low         0.5764299193            0.42357008
## 578              Low         0.1801949699            0.81980503
## 581              Low         0.2337758471            0.76622415
## 585              Low         0.3222395378            0.67776046
## 590              Low         0.7000901095            0.29990989
## 601              Low         0.7965076159            0.20349238
## 602              Low         0.5911455454            0.40885445
## 607              Low         0.5745963409            0.42540366
## 610              Low         0.5598294027            0.44017060
## 618              Low         0.6718083378            0.32819166
## 624              Low         0.3222395378            0.67776046
## 626              Low         0.2300137304            0.76998627
## 627              Low         0.3796756641            0.62032434
## 634              Low         0.6772469161            0.32275308
## 640              Low         0.9826668299            0.01733317
## 642              Low         0.1596173546            0.84038265
## 643              Low         0.6603529106            0.33964709
## 644              Low         0.8722361193            0.12776388
## 645              Low         0.7805565258            0.21944347
## 646              Low         0.7944944044            0.20550560
## 647              Low         0.7214006575            0.27859934
## 652              Low         0.1909579025            0.80904210
## 658              Low         0.5578158104            0.44218419
## 659              Low         0.8677054479            0.13229455
## 660              Low         0.8660831750            0.13391682
## 664              Low         0.3154666895            0.68453331
## 666              Low         0.4094602677            0.59053973
## 667              Low         0.8101180279            0.18988197
## 675              Low         0.6603529106            0.33964709
## 680              Low         0.9641799867            0.03582001
## 681              Low         0.8594185600            0.14058144
## 687              Low         0.8801008895            0.11989911
## 694              Low         0.7275971055            0.27240289
## 697              Low         0.5638293124            0.43617069
## 701              Low         0.3818211000            0.61817890
## 705              Low         0.9049764620            0.09502354
## 707              Low         0.6651475934            0.33485241
## 710              Low         0.7664857792            0.23351422
## 716              Low         0.8679648226            0.13203518
## 719              Low         0.9391300604            0.06086994
## 720              Low         0.9749949067            0.02500509
## 725              Low         0.9760911253            0.02390887
## 727              Low         0.3818211000            0.61817890
## 730              Low         0.5141175113            0.48588249
## 738              Low         0.8712602625            0.12873974
## 745              Low         0.7303903111            0.26960969
## 748              Low         0.8143622451            0.18563775
## 751              Low         0.9236735195            0.07632648
## 756              Low         0.6594624920            0.34053751
## 766              Low         0.7257122288            0.27428777
## 769              Low         0.6219413487            0.37805865
## 783              Low         0.8773414681            0.12265853
## 785              Low         0.9119616001            0.08803840
## 790              Low         0.8742200814            0.12577992
## 793              Low         0.8773414681            0.12265853
## 795              Low         0.9873215475            0.01267845
## 796              Low         0.9797901295            0.02020987
## 797              Low         0.5245149624            0.47548504
## 801              Low         0.7819966694            0.21800333
## 811              Low         0.5830613359            0.41693866
## 812              Low         0.9209125651            0.07908743
## 815              Low         0.9579291216            0.04207088
## 816              Low         0.7563111238            0.24368888
## 817              Low         0.9554192278            0.04458077
## 824              Low         0.8910529193            0.10894708
## 825              Low         0.8910529193            0.10894708
## 826              Low         0.8910529193            0.10894708
## 830              Low         0.9206758207            0.07932418
## 837              Low         0.9458135660            0.05418643
## 838              Low         0.7563111238            0.24368888
## 844              Low         0.9344794385            0.06552056
## 845              Low         0.9788289697            0.02117103
## 847              Low         0.9342189025            0.06578110
## 850              Low         0.9042545052            0.09574549
## 852              Low         0.9162961404            0.08370386
## 853              Low         0.9162961404            0.08370386
## 861              Low         0.9270249964            0.07297500
## 868              Low         0.9821061387            0.01789386
## 874              Low         0.9367252603            0.06327474
## 879             High         0.1033547908            0.89664521
## 895             High         0.0101263144            0.98987369
## 899             High         0.0009470188            0.99905298
## 903             High         0.0101263144            0.98987369
## 917             High         0.0647560019            0.93524400
## 927             High         0.0211968888            0.97880311
## 929             High         0.1060669672            0.89393303
## 931             High         0.0211968888            0.97880311
## 933             High         0.3554801761            0.64451982
## 944             High         0.0433148853            0.95668511
## 947             High         0.0398746182            0.96012538
## 949             High         0.1877588381            0.81224116
## 953             High         0.0233211466            0.97667885
## 958             High         0.5637319742            0.43626803
## 961             High         0.0150665037            0.98493350
## 963             High         0.1129960710            0.88700393
## 964             High         0.0687190029            0.93128100
## 973             High         0.0532886213            0.94671138
## 976             High         0.0264753239            0.97352468
## 977             High         0.1682281293            0.83177187
## 980             High         0.2904095993            0.70959040
## 983             High         0.6727220240            0.32727798
## 984             High         0.1682281293            0.83177187
## 986             High         0.1089806628            0.89101934
## 989             High         0.2022144660            0.79778553
## 991             High         0.0240809741            0.97591903
## 996             High         0.0333823806            0.96661762
## 997             High         0.4975165837            0.50248342
## 999             High         0.0465501013            0.95344990
## 1000            High         0.0565000618            0.94349994
## 1003            High         0.0259764904            0.97402351
## 1008            High         0.1025560666            0.89744393
## 1009            High         0.4656557701            0.53434423
## 1014            High         0.0413949131            0.95860509
## 1015            High         0.4555939327            0.54440607
## 1040            High         0.1848342215            0.81516578
## 1042            High         0.3372214149            0.66277859
## 1043            High         0.7410621250            0.25893787
## 1050            High         0.1112632945            0.88873671
## 1052            High         0.1738234646            0.82617654
## 1056            High         0.3235717630            0.67642824
## 1070            High         0.5311755155            0.46882448
## 1073            High         0.5151860040            0.48481400
## 1074            High         0.1702113670            0.82978863
## 1079            High         0.4374952373            0.56250476
## 1080            High         0.6262738215            0.37372618
## 1085            High         0.0855709853            0.91442901
## 1087            High         0.6387004046            0.36129960
## 1096            High         0.8979928982            0.10200710
## 1099            High         0.4266976843            0.57330232
## 1100            High         0.7178153041            0.28218470
## 1102            High         0.0821493064            0.91785069
## 1107             Low         0.4800969559            0.51990304
## 1109             Low         0.6555391529            0.34446085
## 1114             Low         0.5204465731            0.47955343
## 1118             Low         0.3451967905            0.65480321
## 1123             Low         0.4281009860            0.57189901
## 1132             Low         0.8951347736            0.10486523
## 1134             Low         0.5798822776            0.42011772
## 1137             Low         0.3796756641            0.62032434
## 1154             Low         0.3796756641            0.62032434
## 1155             Low         0.5576566154            0.44234338
## 1157             Low         0.7299812013            0.27001880
## 1162             Low         0.6603529106            0.33964709
## 1164             Low         0.1596173546            0.84038265
## 1171             Low         0.9236735195            0.07632648
## 1172             Low         0.5324789498            0.46752105
## 1175             Low         0.6025547990            0.39744520
## 1177             Low         0.7033318635            0.29666814
## 1179             Low         0.8937618468            0.10623815
## 1183             Low         0.2760144412            0.72398556
## 1185             Low         0.9280768920            0.07192311
## 1189             Low         0.8086497299            0.19135027
## 1211             Low         0.8354424008            0.16455760
## 1218             Low         0.9892374352            0.01076256
## 1224             Low         0.4458631939            0.55413681
## 1225             Low         0.3818211000            0.61817890
## 1227             Low         0.8964097761            0.10359022
## 1232             Low         0.9844700142            0.01552999
## 1235             Low         0.7387527307            0.26124727
## 1238             Low         0.6806068698            0.31939313
## 1240             Low         0.8384203887            0.16157961
## 1241             Low         0.8354424008            0.16455760
## 1248             Low         0.9206758207            0.07932418
## 1258             Low         0.7563111238            0.24368888
## 1261             Low         0.8844033428            0.11559666
## 1263             Low         0.8910529193            0.10894708
## 1269             Low         0.9557472206            0.04425278
## 1270             Low         0.9853749858            0.01462501
## 1271             Low         0.9042545052            0.09574549
## 1272             Low         0.9162961404            0.08370386
## 1280             Low         0.9367252603            0.06327474
## 1286             Low         0.9875508520            0.01244915
## 1287             Low         0.9885358662            0.01146413
## 1289             Low         0.9523876595            0.04761234
## 1290             Low         0.9162961404            0.08370386
## 1291            High         0.1783988428            0.82160116
## 1294            High         0.8086918533            0.19130815
## 1305             Low         0.8712582117            0.12874179
## 1308            High         0.7148932523            0.28510675

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_YJT_Test_ROC <- roc(response = LR_YJT_Test$LR_YJT_Observed,
             predictor = LR_YJT_Test$LR_YJT_Predicted.High,
             levels = rev(levels(LR_YJT_Test$LR_YJT_Observed)))

(LR_YJT_Test_ROCCurveAUC <- auc(LR_YJT_Test_ROC)[1])

## [1] 0.8906181

1.5.4 Logistic Regression With Exponential Transformation (LR_ET)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Exponential Transformation applies a new family of exponential distributions that can be used without restrictions, providing a good alternative for other power transformation families. While there are no restrictions in terms of the applicable values, the method’s effectiveness is sensitive to the quality of the data as it needs to assume a common mean and estimate the transformation parameter by directly maximizing the likelihood.

[A] The logistic regression model from the stats package was implemented through the caret package with exponential transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.88053

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89013

Code Chunk | Output

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
Transform_Exponential <- preProcess(PMA_PreModelling_Train_LR, method = c("expoTrans"))
PMA_PreModelling_Train_LR_ET <- predict(Transform_Exponential, PMA_PreModelling_Train_LR.Numeric)
PMA_PreModelling_Train_LR_ET$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_ET[,sapply(PMA_PreModelling_Train_LR_ET, is.numeric)],
            y = PMA_PreModelling_Train_LR_ET$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_ET Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_ET[,sapply(PMA_PreModelling_Train_LR_ET, is.numeric)], 
            y = PMA_PreModelling_Train_LR_ET$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_ET Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_ET$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_ET$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_ET_Tune <- train(x = PMA_PreModelling_Train_LR_ET[,!names(PMA_PreModelling_Train_LR_ET) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_ET$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_ET_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec    
##   0.8805333  0.7187154  0.830225

LR_ET_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##            6.2403             1.4785             0.1379            -0.7367  
##         NumCarbon  
##           -0.1887  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 811.6     AIC: 821.6

LR_ET_Tune$results

##   parameter       ROC      Sens     Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8805333 0.7187154 0.830225 0.03607516 0.07895841 0.06335689

(LR_ET_Train_ROCCurveAUC <- LR_ET_Tune$results$ROC)

## [1] 0.8805333

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_ET_VarImp <- varImp(LR_ET_Tune, scale = TRUE)
plot(LR_ET_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_ET <- predict(Transform_Exponential, PMA_PreModelling_Test_LR.Numeric)
PMA_PreModelling_Test_LR_ET$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_ET[,sapply(PMA_PreModelling_Test_LR_ET, is.numeric)],
            y = PMA_PreModelling_Test_LR_ET$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_ET Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_ET[,sapply(PMA_PreModelling_Test_LR_ET, is.numeric)], 
            y = PMA_PreModelling_Test_LR_ET$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_ET Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_ET$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_ET_Test <- data.frame(LR_ET_Observed = PMA_PreModelling_Test_LR_ET$Log_Solubility_Class,
                      LR_ET_Predicted = predict(LR_ET_Tune,
                      PMA_PreModelling_Test_LR_ET[,!names(PMA_PreModelling_Test_LR_ET) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_ET_Test

##      LR_ET_Observed LR_ET_Predicted.Low LR_ET_Predicted.High
## 20             High         0.013752867           0.98624713
## 21             High         0.009194143           0.99080586
## 23             High         0.044071687           0.95592831
## 25             High         0.015356098           0.98464390
## 28             High         0.062116193           0.93788381
## 31             High         0.030063697           0.96993630
## 32             High         0.055492297           0.94450770
## 33             High         0.081840730           0.91815927
## 34             High         0.081840730           0.91815927
## 37             High         0.362323531           0.63767647
## 38             High         0.362323531           0.63767647
## 42             High         0.529862825           0.47013717
## 49             High         0.175456910           0.82454309
## 54             High         0.027319430           0.97268057
## 55             High         0.022068424           0.97793158
## 58             High         0.443511260           0.55648874
## 60             High         0.082693666           0.91730633
## 61             High         0.081840730           0.91815927
## 65             High         0.048911337           0.95108866
## 69             High         0.533781158           0.46621884
## 73             High         0.032939911           0.96706009
## 86             High         0.067654064           0.93234594
## 90             High         0.046903055           0.95309695
## 91             High         0.016987566           0.98301243
## 93             High         0.046903055           0.95309695
## 96             High         0.016987566           0.98301243
## 98             High         0.053351489           0.94664851
## 100            High         0.066118585           0.93388142
## 104            High         0.892159099           0.10784090
## 112            High         0.226736762           0.77326324
## 115            High         0.880645514           0.11935449
## 119            High         0.153976304           0.84602370
## 128            High         0.153976304           0.84602370
## 130            High         0.023851968           0.97614803
## 139            High         0.023851968           0.97614803
## 143            High         0.038552905           0.96144709
## 145            High         0.114662957           0.88533704
## 146            High         0.153976304           0.84602370
## 149            High         0.152805307           0.84719469
## 150            High         0.133318136           0.86668186
## 152            High         0.081840730           0.91815927
## 157            High         0.513142137           0.48685786
## 161            High         0.245051293           0.75494871
## 162            High         0.010303091           0.98969691
## 166            High         0.528097208           0.47190279
## 167            High         0.164575958           0.83542404
## 173            High         0.116830681           0.88316932
## 176            High         0.153976304           0.84602370
## 182            High         0.020554918           0.97944508
## 187            High         0.049933006           0.95006699
## 190            High         0.020110335           0.97988967
## 194            High         0.023549684           0.97645032
## 195            High         0.178797120           0.82120288
## 201            High         0.076255775           0.92374422
## 207            High         0.117776492           0.88222351
## 208            High         0.511302373           0.48869763
## 215            High         0.030063697           0.96993630
## 222            High         0.244990513           0.75500949
## 224            High         0.182053122           0.81794688
## 231            High         0.749586694           0.25041331
## 236            High         0.097554294           0.90244571
## 237            High         0.041892008           0.95810799
## 240            High         0.185612595           0.81438741
## 243            High         0.076255775           0.92374422
## 248            High         0.178797120           0.82120288
## 251            High         0.775565527           0.22443447
## 256            High         0.398539794           0.60146021
## 258            High         0.169197351           0.83080265
## 262            High         0.511302373           0.48869763
## 266            High         0.452920897           0.54707910
## 272            High         0.570995637           0.42900436
## 280            High         0.296779366           0.70322063
## 283            High         0.443886349           0.55611365
## 286            High         0.435270439           0.56472956
## 287            High         0.168888318           0.83111168
## 289            High         0.099729418           0.90027058
## 290            High         0.316994422           0.68300558
## 298            High         0.318073613           0.68192639
## 305            High         0.294523101           0.70547690
## 306            High         0.222812248           0.77718775
## 312            High         0.101828758           0.89817124
## 320            High         0.314342671           0.68565733
## 325            High         0.176122072           0.82387793
## 332            High         0.058470118           0.94152988
## 333            High         0.457755060           0.54224494
## 335            High         0.318073613           0.68192639
## 339            High         0.772723681           0.22727632
## 346            High         0.347449045           0.65255096
## 347            High         0.044973673           0.95502633
## 350            High         0.382267109           0.61773289
## 353            High         0.349013643           0.65098636
## 358            High         0.321275416           0.67872458
## 365            High         0.224247273           0.77575273
## 367            High         0.199038406           0.80096159
## 370            High         0.048269146           0.95173085
## 379            High         0.128343126           0.87165687
## 386            High         0.362492091           0.63750791
## 394            High         0.522815273           0.47718473
## 396            High         0.130957795           0.86904220
## 400            High         0.041892008           0.95810799
## 404            High         0.038322544           0.96167746
## 405            High         0.626352870           0.37364713
## 413            High         0.131232803           0.86876720
## 415            High         0.371596531           0.62840347
## 417            High         0.212807558           0.78719244
## 418            High         0.441346753           0.55865325
## 423            High         0.306995054           0.69300495
## 434            High         0.283812634           0.71618737
## 437            High         0.274034772           0.72596523
## 440            High         0.378573196           0.62142680
## 449            High         0.362681763           0.63731824
## 450            High         0.273348151           0.72665185
## 457            High         0.362681763           0.63731824
## 467            High         0.325012206           0.67498779
## 469            High         0.221752671           0.77824733
## 474            High         0.936674227           0.06332577
## 475            High         0.904392341           0.09560766
## 485            High         0.132581065           0.86741894
## 504             Low         0.288558015           0.71144198
## 511             Low         0.780430879           0.21956912
## 512             Low         0.410836585           0.58916341
## 517             Low         0.082425749           0.91757425
## 519             Low         0.689505392           0.31049461
## 520             Low         0.075648339           0.92435166
## 522             Low         0.940550594           0.05944941
## 527             Low         0.635155993           0.36484401
## 528             Low         0.295989467           0.70401053
## 529             Low         0.301900165           0.69809983
## 537             Low         0.124694956           0.87530504
## 540             Low         0.924323450           0.07567655
## 541             Low         0.545190763           0.45480924
## 547             Low         0.920545497           0.07945450
## 550             Low         0.769867948           0.23013205
## 555             Low         0.500850846           0.49914915
## 564             Low         0.037015067           0.96298493
## 570             Low         0.354179129           0.64582087
## 573             Low         0.238387555           0.76161244
## 575             Low         0.545190763           0.45480924
## 578             Low         0.188629199           0.81137080
## 581             Low         0.238387555           0.76161244
## 585             Low         0.323229930           0.67677007
## 590             Low         0.723611110           0.27638889
## 601             Low         0.797056033           0.20294397
## 602             Low         0.613964299           0.38603570
## 607             Low         0.592073516           0.40792648
## 610             Low         0.577538414           0.42246159
## 618             Low         0.685443653           0.31455635
## 624             Low         0.323229930           0.67677007
## 626             Low         0.241740192           0.75825981
## 627             Low         0.357226702           0.64277330
## 634             Low         0.660110865           0.33988913
## 640             Low         0.982153120           0.01784688
## 642             Low         0.168057998           0.83194200
## 643             Low         0.628905923           0.37109408
## 644             Low         0.867026175           0.13297383
## 645             Low         0.773312713           0.22668729
## 646             Low         0.770707712           0.22929229
## 647             Low         0.714109400           0.28589060
## 652             Low         0.191323910           0.80867609
## 658             Low         0.575949760           0.42405024
## 659             Low         0.851385294           0.14861471
## 660             Low         0.880217375           0.11978263
## 664             Low         0.319822683           0.68017732
## 666             Low         0.416063033           0.58393697
## 667             Low         0.827447505           0.17255250
## 675             Low         0.628905923           0.37109408
## 680             Low         0.964633230           0.03536677
## 681             Low         0.874768134           0.12523187
## 687             Low         0.866602938           0.13339706
## 694             Low         0.748302599           0.25169740
## 697             Low         0.559681179           0.44031882
## 701             Low         0.359811881           0.64018812
## 705             Low         0.912000344           0.08799966
## 707             Low         0.700044807           0.29995519
## 710             Low         0.743622998           0.25637700
## 716             Low         0.885887800           0.11411220
## 719             Low         0.939669824           0.06033018
## 720             Low         0.974804926           0.02519507
## 725             Low         0.976039979           0.02396002
## 727             Low         0.359811881           0.64018812
## 730             Low         0.504435382           0.49556462
## 738             Low         0.859253410           0.14074659
## 745             Low         0.702405944           0.29759406
## 748             Low         0.808544204           0.19145580
## 751             Low         0.933147457           0.06685254
## 756             Low         0.685429835           0.31457016
## 766             Low         0.751514843           0.24848516
## 769             Low         0.602517547           0.39748245
## 783             Low         0.869255200           0.13074480
## 785             Low         0.902579681           0.09742032
## 790             Low         0.886834280           0.11316572
## 793             Low         0.869255200           0.13074480
## 795             Low         0.986490055           0.01350995
## 796             Low         0.979675106           0.02032489
## 797             Low         0.555989916           0.44401008
## 801             Low         0.775047694           0.22495231
## 811             Low         0.556377102           0.44362290
## 812             Low         0.929865865           0.07013413
## 815             Low         0.957749970           0.04225003
## 816             Low         0.758028058           0.24197194
## 817             Low         0.951236951           0.04876305
## 824             Low         0.886411056           0.11358894
## 825             Low         0.886411056           0.11358894
## 826             Low         0.886411056           0.11358894
## 830             Low         0.912117000           0.08788300
## 837             Low         0.940840521           0.05915948
## 838             Low         0.758028058           0.24197194
## 844             Low         0.927974648           0.07202535
## 845             Low         0.976894550           0.02310545
## 847             Low         0.933393677           0.06660632
## 850             Low         0.902116294           0.09788371
## 852             Low         0.915857107           0.08414289
## 853             Low         0.915857107           0.08414289
## 861             Low         0.927636110           0.07236389
## 868             Low         0.980340828           0.01965917
## 874             Low         0.937852871           0.06214713
## 879            High         0.113209880           0.88679012
## 895            High         0.014930232           0.98506977
## 899            High         0.001859651           0.99814035
## 903            High         0.014930232           0.98506977
## 917            High         0.056418013           0.94358199
## 927            High         0.027319430           0.97268057
## 929            High         0.116830681           0.88316932
## 931            High         0.027319430           0.97268057
## 933            High         0.362953760           0.63704624
## 944            High         0.043101463           0.95689854
## 947            High         0.046903055           0.95309695
## 949            High         0.184608445           0.81539155
## 953            High         0.021584412           0.97841559
## 958            High         0.536045815           0.46395419
## 961            High         0.015768127           0.98423187
## 963            High         0.119476478           0.88052352
## 964            High         0.076255775           0.92374422
## 973            High         0.049972348           0.95002765
## 976            High         0.037015067           0.96298493
## 977            High         0.178797120           0.82120288
## 980            High         0.275965540           0.72403446
## 983            High         0.650018484           0.34998152
## 984            High         0.178797120           0.82120288
## 986            High         0.101828758           0.89817124
## 989            High         0.202228735           0.79777127
## 991            High         0.023066118           0.97693388
## 996            High         0.050382284           0.94961772
## 997            High         0.470933547           0.52906645
## 999            High         0.044071687           0.95592831
## 1000           High         0.051069687           0.94893031
## 1003           High         0.041892008           0.95810799
## 1008           High         0.099376550           0.90062345
## 1009           High         0.443886349           0.55611365
## 1014           High         0.041232631           0.95876737
## 1015           High         0.470359921           0.52964008
## 1040           High         0.177580468           0.82241953
## 1042           High         0.339492531           0.66050747
## 1043           High         0.730538815           0.26946119
## 1050           High         0.104620962           0.89537904
## 1052           High         0.180674837           0.81932516
## 1056           High         0.333657886           0.66634211
## 1070           High         0.555232608           0.44476739
## 1073           High         0.486744863           0.51325514
## 1074           High         0.177106183           0.82289382
## 1079           High         0.416066054           0.58393395
## 1080           High         0.599211550           0.40078845
## 1085           High         0.082405863           0.91759414
## 1087           High         0.657183051           0.34281695
## 1096           High         0.909565322           0.09043468
## 1099           High         0.440418289           0.55958171
## 1100           High         0.699230090           0.30076991
## 1102           High         0.071892116           0.92810788
## 1107            Low         0.454403473           0.54559653
## 1109            Low         0.675224568           0.32477543
## 1114            Low         0.491991507           0.50800849
## 1118            Low         0.356946854           0.64305315
## 1123            Low         0.432009924           0.56799008
## 1132            Low         0.886272790           0.11372721
## 1134            Low         0.603826999           0.39617300
## 1137            Low         0.357226702           0.64277330
## 1154            Low         0.357226702           0.64277330
## 1155            Low         0.582463212           0.41753679
## 1157            Low         0.765117005           0.23488300
## 1162            Low         0.628905923           0.37109408
## 1164            Low         0.168057998           0.83194200
## 1171            Low         0.933147457           0.06685254
## 1172            Low         0.513142137           0.48685786
## 1175            Low         0.627703257           0.37229674
## 1177            Low         0.687471315           0.31252868
## 1179            Low         0.902338418           0.09766158
## 1183            Low         0.268533199           0.73146680
## 1185            Low         0.934234085           0.06576592
## 1189            Low         0.820187858           0.17981214
## 1211            Low         0.816488095           0.18351190
## 1218            Low         0.987709192           0.01229081
## 1224            Low         0.418627487           0.58137251
## 1225            Low         0.359811881           0.64018812
## 1227            Low         0.906940352           0.09305965
## 1232            Low         0.983937916           0.01606208
## 1235            Low         0.755569564           0.24443044
## 1238            Low         0.713703436           0.28629656
## 1240            Low         0.855212853           0.14478715
## 1241            Low         0.816488095           0.18351190
## 1248            Low         0.912117000           0.08788300
## 1258            Low         0.758028058           0.24197194
## 1261            Low         0.886758010           0.11324199
## 1263            Low         0.886411056           0.11358894
## 1269            Low         0.955630849           0.04436915
## 1270            Low         0.985258898           0.01474110
## 1271            Low         0.902116294           0.09788371
## 1272            Low         0.915857107           0.08414289
## 1280            Low         0.937852871           0.06214713
## 1286            Low         0.986020488           0.01397951
## 1287            Low         0.987076206           0.01292379
## 1289            Low         0.953597078           0.04640292
## 1290            Low         0.915857107           0.08414289
## 1291           High         0.175452898           0.82454710
## 1294           High         0.799908047           0.20009195
## 1305            Low         0.880239111           0.11976089
## 1308           High         0.727385856           0.27261414

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_ET_Test_ROC <- roc(response = LR_ET_Test$LR_ET_Observed,
             predictor = LR_ET_Test$LR_ET_Predicted.High,
             levels = rev(levels(LR_ET_Test$LR_ET_Observed)))

(LR_ET_Test_ROCCurveAUC <- auc(LR_ET_Test_ROC)[1])

## [1] 0.890133

1.5.5 Logistic Regression With Inverse Hyperbolic Sine Transformation (LR_IHST)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Inverse Hyperbolic Sine Transformation behaves similarly to a log transformation for positive values. The inverse hyperbolic since function however has the added advantage of remaining defined for zeroes and negative values.

[A] The logistic regression model from the stats package was implemented through the caret package with inverse hyperbolic sine transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.87107

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.87998

Code Chunk | Output

##################################
# Applying inverse hyperbolic sine function
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  PMA_PreModelling_Train[,i] <- log(PMA_PreModelling_Train[,i]+(((PMA_PreModelling_Train[,i])^2)+1)^(1/2))
}

for (i in 1:(ncol(PMA_PreModelling_Test)-1)){
  PMA_PreModelling_Test[,i] <- log(PMA_PreModelling_Test[,i]+(((PMA_PreModelling_Test[,i])^2)+1)^(1/2))
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
PMA_PreModelling_Train_LR_IHST <- PMA_PreModelling_Train_LR
PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_IHST[,sapply(PMA_PreModelling_Train_LR_IHST, is.numeric)],
            y = PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_IHST Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_IHST[,sapply(PMA_PreModelling_Train_LR_IHST, is.numeric)], 
            y = PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_IHST Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_IHST_Tune <- train(x = PMA_PreModelling_Train_LR_IHST[,!names(PMA_PreModelling_Train_LR_IHST) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_IHST$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_IHST_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8710722  0.7374862  0.8073657

LR_IHST_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##            10.721              1.364              1.208             -3.671  
##         NumCarbon  
##            -1.128  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 836   AIC: 846

LR_IHST_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD    SpecSD
## 1      none 0.8710722 0.7374862 0.8073657 0.04019999 0.06659904 0.0606913

(LR_IHST_Train_ROCCurveAUC <- LR_IHST_Tune$results$ROC)

## [1] 0.8710722

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_IHST_VarImp <- varImp(LR_IHST_Tune, scale = TRUE)
plot(LR_IHST_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_IHST <- PMA_PreModelling_Test_LR
PMA_PreModelling_Test_LR_IHST$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_IHST[,sapply(PMA_PreModelling_Test_LR_IHST, is.numeric)],
            y = PMA_PreModelling_Test_LR_IHST$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_IHST Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_IHST[,sapply(PMA_PreModelling_Test_LR_IHST, is.numeric)], 
            y = PMA_PreModelling_Test_LR_IHST$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_IHST Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_IHST$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_IHST_Test <- data.frame(LR_IHST_Observed = PMA_PreModelling_Test_LR_IHST$Log_Solubility_Class,
                      LR_IHST_Predicted = predict(LR_IHST_Tune,
                      PMA_PreModelling_Test_LR_IHST[,!names(PMA_PreModelling_Test_LR_IHST) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_IHST_Test

##      LR_IHST_Observed LR_IHST_Predicted.Low LR_IHST_Predicted.High
## 20               High          0.0099300682             0.99006993
## 21               High          0.0070843515             0.99291565
## 23               High          0.0578716298             0.94212837
## 25               High          0.0182500257             0.98174997
## 28               High          0.0801383023             0.91986170
## 31               High          0.0150075039             0.98499250
## 32               High          0.0292220184             0.97077798
## 33               High          0.0524819225             0.94751808
## 34               High          0.0524819225             0.94751808
## 37               High          0.3152799614             0.68472004
## 38               High          0.3152799614             0.68472004
## 42               High          0.6136673812             0.38633262
## 49               High          0.1598231577             0.84017684
## 54               High          0.0356333544             0.96436665
## 55               High          0.0219177360             0.97808226
## 58               High          0.4549805179             0.54501948
## 60               High          0.1172175407             0.88278246
## 61               High          0.0524819225             0.94751808
## 65               High          0.0609565698             0.93904343
## 69               High          0.5374744295             0.46252557
## 73               High          0.0182159523             0.98178405
## 86               High          0.0487705078             0.95122949
## 90               High          0.0693209815             0.93067902
## 91               High          0.0121654002             0.98783460
## 93               High          0.0693209815             0.93067902
## 96               High          0.0121654002             0.98783460
## 98               High          0.0775264455             0.92247355
## 100              High          0.0904961248             0.90950388
## 104              High          0.8993420391             0.10065796
## 112              High          0.1883413325             0.81165867
## 115              High          0.8526100741             0.14738993
## 119              High          0.1405876713             0.85941233
## 128              High          0.1405876713             0.85941233
## 130              High          0.0236487683             0.97635123
## 139              High          0.0236487683             0.97635123
## 143              High          0.0520388523             0.94796115
## 145              High          0.0793526698             0.92064733
## 146              High          0.1405876713             0.85941233
## 149              High          0.2122457494             0.78775425
## 150              High          0.1009300825             0.89906992
## 152              High          0.0524819225             0.94751808
## 157              High          0.5269860724             0.47301393
## 161              High          0.3337627789             0.66623722
## 162              High          0.0062680045             0.99373200
## 166              High          0.5354668322             0.46453317
## 167              High          0.1260047503             0.87399525
## 173              High          0.0887542160             0.91124578
## 176              High          0.1405876713             0.85941233
## 182              High          0.0061870117             0.99381299
## 187              High          0.0576049610             0.94239504
## 190              High          0.0156622796             0.98433772
## 194              High          0.0105820641             0.98941794
## 195              High          0.2551363208             0.74486368
## 201              High          0.1185994482             0.88140055
## 207              High          0.1831137154             0.81688628
## 208              High          0.4561448837             0.54385512
## 215              High          0.0150075039             0.98499250
## 222              High          0.3377831981             0.66221680
## 224              High          0.2291389419             0.77086106
## 231              High          0.7081685175             0.29183148
## 236              High          0.1339363550             0.86606365
## 237              High          0.0201223973             0.97987760
## 240              High          0.1926954630             0.80730454
## 243              High          0.1185994482             0.88140055
## 248              High          0.2551363208             0.74486368
## 251              High          0.8071616550             0.19283835
## 256              High          0.4973558109             0.50264419
## 258              High          0.2393589007             0.76064110
## 262              High          0.4561448837             0.54385512
## 266              High          0.5304409692             0.46955903
## 272              High          0.5720104543             0.42798955
## 280              High          0.3631225228             0.63687748
## 283              High          0.4275080021             0.57249200
## 286              High          0.5141733831             0.48582662
## 287              High          0.2435576257             0.75644237
## 289              High          0.0848641421             0.91513586
## 290              High          0.3819373209             0.61806268
## 298              High          0.3093894899             0.69061051
## 305              High          0.3858465603             0.61415344
## 306              High          0.2315157545             0.76848425
## 312              High          0.1353064070             0.86469359
## 320              High          0.2745723275             0.72542767
## 325              High          0.2358766959             0.76412330
## 332              High          0.0836961949             0.91630381
## 333              High          0.4080892379             0.59191076
## 335              High          0.3093894899             0.69061051
## 339              High          0.7128384446             0.28716156
## 346              High          0.4472904099             0.55270959
## 347              High          0.0300015916             0.96999841
## 350              High          0.3449964094             0.65500359
## 353              High          0.4462698817             0.55373012
## 358              High          0.4225671930             0.57743281
## 365              High          0.3039458499             0.69605415
## 367              High          0.2495985988             0.75040140
## 370              High          0.0427492613             0.95725074
## 379              High          0.1399298135             0.86007019
## 386              High          0.2689183845             0.73108162
## 394              High          0.6069330289             0.39306697
## 396              High          0.1774691742             0.82253083
## 400              High          0.0201223973             0.97987760
## 404              High          0.0221953778             0.97780462
## 405              High          0.5922002466             0.40779975
## 413              High          0.1989446496             0.80105535
## 415              High          0.4673768642             0.53262314
## 417              High          0.1389459796             0.86105402
## 418              High          0.5397594210             0.46024058
## 423              High          0.2609195742             0.73908043
## 434              High          0.3784020245             0.62159798
## 437              High          0.2626702786             0.73732972
## 440              High          0.3698931740             0.63010683
## 449              High          0.4326697948             0.56733021
## 450              High          0.3544261568             0.64557384
## 457              High          0.4326697948             0.56733021
## 467              High          0.3153110496             0.68468895
## 469              High          0.1527762476             0.84722375
## 474              High          0.9075381438             0.09246186
## 475              High          0.8765506823             0.12344932
## 485              High          0.1579281927             0.84207181
## 504               Low          0.2444521260             0.75554787
## 511               Low          0.5871849577             0.41281504
## 512               Low          0.5018320598             0.49816794
## 517               Low          0.0668902556             0.93310974
## 519               Low          0.7251116360             0.27488836
## 520               Low          0.0482164150             0.95178359
## 522               Low          0.9083535976             0.09164640
## 527               Low          0.6995000622             0.30049994
## 528               Low          0.2166802349             0.78331977
## 529               Low          0.3803802257             0.61961977
## 537               Low          0.0944886922             0.90551131
## 540               Low          0.8937605539             0.10623945
## 541               Low          0.4523082593             0.54769174
## 547               Low          0.8943000868             0.10569991
## 550               Low          0.5168207177             0.48317928
## 555               Low          0.5751405754             0.42485942
## 564               Low          0.0210364403             0.97896356
## 570               Low          0.4411424281             0.55885757
## 573               Low          0.3206488297             0.67935117
## 575               Low          0.4523082593             0.54769174
## 578               Low          0.1228104528             0.87718955
## 581               Low          0.3206488297             0.67935117
## 585               Low          0.4095958799             0.59040412
## 590               Low          0.7649838590             0.23501614
## 601               Low          0.7912963243             0.20870368
## 602               Low          0.6719001646             0.32809984
## 607               Low          0.6636629369             0.33633706
## 610               Low          0.6521049110             0.34789509
## 618               Low          0.7096458155             0.29035418
## 624               Low          0.4095958799             0.59040412
## 626               Low          0.3435856801             0.65641432
## 627               Low          0.3417805970             0.65821940
## 634               Low          0.5277070848             0.47229292
## 640               Low          0.9701320053             0.02986799
## 642               Low          0.1097637588             0.89023624
## 643               Low          0.5390069187             0.46099308
## 644               Low          0.8335923318             0.16640767
## 645               Low          0.7520869618             0.24791304
## 646               Low          0.6851967239             0.31480328
## 647               Low          0.6972416523             0.30275835
## 652               Low          0.1601891216             0.83981088
## 658               Low          0.6323549805             0.36764502
## 659               Low          0.7784373668             0.22156263
## 660               Low          0.8882344180             0.11176558
## 664               Low          0.3808059125             0.61919409
## 666               Low          0.4848211779             0.51517882
## 667               Low          0.8424237841             0.15757622
## 675               Low          0.5390069187             0.46099308
## 680               Low          0.9646714157             0.03532858
## 681               Low          0.8853530190             0.11464698
## 687               Low          0.7979014677             0.20209853
## 694               Low          0.7863428810             0.21365712
## 697               Low          0.5780038485             0.42199615
## 701               Low          0.2845925670             0.71540743
## 705               Low          0.9200100288             0.07998997
## 707               Low          0.7455518887             0.25444811
## 710               Low          0.6831664789             0.31683352
## 716               Low          0.8903703029             0.10962970
## 719               Low          0.9144792544             0.08552075
## 720               Low          0.9537538416             0.04624616
## 725               Low          0.9577964846             0.04220352
## 727               Low          0.2845925670             0.71540743
## 730               Low          0.5180529588             0.48194704
## 738               Low          0.8051588342             0.19484117
## 745               Low          0.6164473390             0.38355266
## 748               Low          0.7636704477             0.23632955
## 751               Low          0.9320117380             0.06798826
## 756               Low          0.7357943834             0.26420562
## 766               Low          0.7940349873             0.20596501
## 769               Low          0.6140644801             0.38593552
## 783               Low          0.8278040363             0.17219596
## 785               Low          0.8433731544             0.15662685
## 790               Low          0.8939502894             0.10604971
## 793               Low          0.8278040363             0.17219596
## 795               Low          0.9757396053             0.02426039
## 796               Low          0.9631790196             0.03682098
## 797               Low          0.6283445387             0.37165546
## 801               Low          0.7663863103             0.23361369
## 811               Low          0.5728956904             0.42710431
## 812               Low          0.9288809270             0.07111907
## 815               Low          0.9245343078             0.07546569
## 816               Low          0.7513710061             0.24862899
## 817               Low          0.9110816646             0.08891834
## 824               Low          0.8539043619             0.14609564
## 825               Low          0.8539043619             0.14609564
## 826               Low          0.8539043619             0.14609564
## 830               Low          0.8557141199             0.14428588
## 837               Low          0.8960597684             0.10394023
## 838               Low          0.7513710061             0.24862899
## 844               Low          0.8776431746             0.12235683
## 845               Low          0.9529465696             0.04705343
## 847               Low          0.9085970121             0.09140299
## 850               Low          0.8755465620             0.12445344
## 852               Low          0.8934471700             0.10655283
## 853               Low          0.8934471700             0.10655283
## 861               Low          0.9082923526             0.09170765
## 868               Low          0.9589272001             0.04107280
## 874               Low          0.9207968258             0.07920317
## 879              High          0.0946843934             0.90531561
## 895              High          0.0154204755             0.98457952
## 899              High          0.0001948401             0.99980516
## 903              High          0.0154204755             0.98457952
## 917              High          0.0596706125             0.94032939
## 927              High          0.0356333544             0.96436665
## 929              High          0.0887542160             0.91124578
## 931              High          0.0356333544             0.96436665
## 933              High          0.4369472072             0.56305279
## 944              High          0.0614810203             0.93851898
## 947              High          0.0693209815             0.93067902
## 949              High          0.2464938138             0.75350619
## 953              High          0.0184962609             0.98150374
## 958              High          0.4853511053             0.51464889
## 961              High          0.0213531141             0.97864689
## 963              High          0.1129852773             0.88701472
## 964              High          0.1185994482             0.88140055
## 973              High          0.0653591674             0.93464083
## 976              High          0.0210364403             0.97896356
## 977              High          0.2551363208             0.74486368
## 980              High          0.2741358706             0.72586413
## 983              High          0.6342816182             0.36571838
## 984              High          0.2551363208             0.74486368
## 986              High          0.1353064070             0.86469359
## 989              High          0.1686883204             0.83131168
## 991              High          0.0281518759             0.97184812
## 996              High          0.0254700687             0.97452993
## 997              High          0.4490879688             0.55091203
## 999              High          0.0578716298             0.94212837
## 1000             High          0.0600594201             0.93994058
## 1003             High          0.0201223973             0.97987760
## 1008             High          0.1412171683             0.85878283
## 1009             High          0.4275080021             0.57249200
## 1014             High          0.0667172124             0.93328279
## 1015             High          0.5346752396             0.46532476
## 1040             High          0.2265614990             0.77343850
## 1042             High          0.4024945490             0.59750545
## 1043             High          0.7138824116             0.28611759
## 1050             High          0.1370627240             0.86293728
## 1052             High          0.2538067860             0.74619321
## 1056             High          0.1534757420             0.84652426
## 1070             High          0.6331959179             0.36680408
## 1073             High          0.5012647881             0.49873521
## 1074             High          0.1370051566             0.86299484
## 1079             High          0.4040490277             0.59595097
## 1080             High          0.5480798185             0.45192018
## 1085             High          0.0730303379             0.92696966
## 1087             High          0.7155593169             0.28444068
## 1096             High          0.9127890547             0.08721095
## 1099             High          0.5132163700             0.48678363
## 1100             High          0.6607261272             0.33927387
## 1102             High          0.0746000948             0.92539991
## 1107              Low          0.3601428127             0.63985719
## 1109              Low          0.7166172165             0.28338278
## 1114              Low          0.3934733313             0.60652667
## 1118              Low          0.4477738160             0.55222618
## 1123              Low          0.4773018833             0.52269812
## 1132              Low          0.8357851715             0.16421483
## 1134              Low          0.6748232479             0.32517675
## 1137              Low          0.3417805970             0.65821940
## 1154              Low          0.3417805970             0.65821940
## 1155              Low          0.6580481264             0.34195187
## 1157              Low          0.7989572984             0.20104270
## 1162              Low          0.5390069187             0.46099308
## 1164              Low          0.1097637588             0.89023624
## 1171              Low          0.9320117380             0.06798826
## 1172              Low          0.5269860724             0.47301393
## 1175              Low          0.6872275852             0.31277241
## 1177              Low          0.6517991969             0.34820080
## 1179              Low          0.9066712497             0.09332875
## 1183              Low          0.1968901528             0.80310985
## 1185              Low          0.9353748426             0.06462516
## 1189              Low          0.8149583318             0.18504167
## 1211              Low          0.7381952599             0.26180474
## 1218              Low          0.9802534524             0.01974655
## 1224              Low          0.4232012115             0.57679879
## 1225              Low          0.2845925670             0.71540743
## 1227              Low          0.9120709631             0.08792904
## 1232              Low          0.9714750094             0.02852499
## 1235              Low          0.7910733456             0.20892665
## 1238              Low          0.7635803175             0.23641968
## 1240              Low          0.8695451356             0.13045486
## 1241              Low          0.7381952599             0.26180474
## 1248              Low          0.8557141199             0.14428588
## 1258              Low          0.7513710061             0.24862899
## 1261              Low          0.8706811127             0.12931889
## 1263              Low          0.8539043619             0.14609564
## 1269              Low          0.9367083202             0.06329168
## 1270              Low          0.9728605279             0.02713947
## 1271              Low          0.8755465620             0.12445344
## 1272              Low          0.8934471700             0.10655283
## 1280              Low          0.9207968258             0.07920317
## 1286              Low          0.9697385070             0.03026149
## 1287              Low          0.9716992194             0.02830078
## 1289              Low          0.9400342381             0.05996576
## 1290              Low          0.8934471700             0.10655283
## 1291             High          0.2366679120             0.76333209
## 1294             High          0.7680428374             0.23195716
## 1305              Low          0.8943028215             0.10569718
## 1308             High          0.7223131307             0.27768687

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_IHST_Test_ROC <- roc(response = LR_IHST_Test$LR_IHST_Observed,
             predictor = LR_IHST_Test$LR_IHST_Predicted.High,
             levels = rev(levels(LR_IHST_Test$LR_IHST_Observed)))

(LR_IHST_Test_ROCCurveAUC <- auc(LR_IHST_Test_ROC)[1])

## [1] 0.8799871

1.5.6 Logistic Regression With Base-10 Logarithm Transformation (LR_LOG10T)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Base-10 Logarithm Transformation applies an algebraic transformation by substituting the variable by its logarithm function-applied form using a base equal to 10. As the logarithm of zero or any negative number is undefined, a constant must be added to move the minimum value of the distribution preferably to 1.

[A] The logistic regression model from the stats package was implemented through the caret package with base-10 logarithm transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.89192

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89882

Code Chunk | Output

##################################
# Applying numerical adjustments
# to eliminate zero values and
# base10-logarithm function
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  PMA_PreModelling_Train[,i] <- PMA_PreModelling_Train[,i] + 1
  PMA_PreModelling_Train[,i] <- log10(PMA_PreModelling_Train[,i])
}

for (i in 1:(ncol(PMA_PreModelling_Test)-1)){
  PMA_PreModelling_Test[,i] <- PMA_PreModelling_Test[,i] + 1
  PMA_PreModelling_Test[,i] <- log10(PMA_PreModelling_Test[,i])
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
PMA_PreModelling_Train_LR_LOG10T <- PMA_PreModelling_Train_LR
PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_LOG10T[,sapply(PMA_PreModelling_Train_LR_LOG10T, is.numeric)],
            y = PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_LOG10T Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_LOG10T[,sapply(PMA_PreModelling_Train_LR_LOG10T, is.numeric)], 
            y = PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_LOG10T Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_LOG10T_Tune <- train(x = PMA_PreModelling_Train_LR_LOG10T[,!names(PMA_PreModelling_Train_LR_LOG10T) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_LOG10T$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_LOG10T_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC       Sens       Spec     
##   0.891921  0.7492802  0.8245646

LR_LOG10T_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##            11.417              2.928              3.279            -14.187  
##         NumCarbon  
##             0.934  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 767.4     AIC: 777.4

LR_LOG10T_Tune$results

##   parameter      ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.891921 0.7492802 0.8245646 0.03393011 0.06711081 0.06719134

(LR_LOG10T_Train_ROCCurveAUC <- LR_LOG10T_Tune$results$ROC)

## [1] 0.891921

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_LOG10T_VarImp <- varImp(LR_LOG10T_Tune, scale = TRUE)
plot(LR_LOG10T_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_LOG10T <- PMA_PreModelling_Test_LR
PMA_PreModelling_Test_LR_LOG10T$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_LOG10T[,sapply(PMA_PreModelling_Test_LR_LOG10T, is.numeric)],
            y = PMA_PreModelling_Test_LR_LOG10T$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_LOG10T Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_LOG10T[,sapply(PMA_PreModelling_Test_LR_LOG10T, is.numeric)], 
            y = PMA_PreModelling_Test_LR_LOG10T$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_LOG10T Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_LOG10T$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_LOG10T_Test <- data.frame(LR_LOG10T_Observed = PMA_PreModelling_Test_LR_LOG10T$Log_Solubility_Class,
                      LR_LOG10T_Predicted = predict(LR_LOG10T_Tune,
                      PMA_PreModelling_Test_LR_LOG10T[,!names(PMA_PreModelling_Test_LR_LOG10T) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_LOG10T_Test

##      LR_LOG10T_Observed LR_LOG10T_Predicted.Low LR_LOG10T_Predicted.High
## 20                 High            0.0114244517             0.9885755483
## 21                 High            0.0080310909             0.9919689091
## 23                 High            0.0354283423             0.9645716577
## 25                 High            0.0113319188             0.9886680812
## 28                 High            0.0603784377             0.9396215623
## 31                 High            0.0126421937             0.9873578063
## 32                 High            0.0204088121             0.9795911879
## 33                 High            0.0401610347             0.9598389653
## 34                 High            0.0401610347             0.9598389653
## 37                 High            0.3543519894             0.6456480106
## 38                 High            0.3543519894             0.6456480106
## 42                 High            0.4516439293             0.5483560707
## 49                 High            0.1326709535             0.8673290465
## 54                 High            0.0127138418             0.9872861582
## 55                 High            0.0171425230             0.9828574770
## 58                 High            0.4508886389             0.5491113611
## 60                 High            0.0458725899             0.9541274101
## 61                 High            0.0401610347             0.9598389653
## 65                 High            0.0422682755             0.9577317245
## 69                 High            0.5360931241             0.4639068759
## 73                 High            0.0313344498             0.9686655502
## 86                 High            0.0379168490             0.9620831510
## 90                 High            0.0246356037             0.9753643963
## 91                 High            0.0112519351             0.9887480649
## 93                 High            0.0246356037             0.9753643963
## 96                 High            0.0112519351             0.9887480649
## 98                 High            0.0281163829             0.9718836171
## 100                High            0.0455991395             0.9544008605
## 104                High            0.8596843413             0.1403156587
## 112                High            0.1899203318             0.8100796682
## 115                High            0.8830092589             0.1169907411
## 119                High            0.1145296081             0.8854703919
## 128                High            0.1145296081             0.8854703919
## 130                High            0.0186594977             0.9813405023
## 139                High            0.0186594977             0.9813405023
## 143                High            0.0369097104             0.9630902896
## 145                High            0.0632876797             0.9367123203
## 146                High            0.1145296081             0.8854703919
## 149                High            0.1330597875             0.8669402125
## 150                High            0.0881581152             0.9118418848
## 152                High            0.0401610347             0.9598389653
## 157                High            0.5491594648             0.4508405352
## 161                High            0.1818965999             0.8181034001
## 162                High            0.0114108307             0.9885891693
## 166                High            0.5306213331             0.4693786669
## 167                High            0.1130629002             0.8869370998
## 173                High            0.0762280843             0.9237719157
## 176                High            0.1145296081             0.8854703919
## 182                High            0.0040896523             0.9959103477
## 187                High            0.0411811702             0.9588188298
## 190                High            0.0232842897             0.9767157103
## 194                High            0.0175444271             0.9824555729
## 195                High            0.1108113031             0.8891886969
## 201                High            0.0436230271             0.9563769729
## 207                High            0.0716468222             0.9283531778
## 208                High            0.5395740091             0.4604259909
## 215                High            0.0126421937             0.9873578063
## 222                High            0.1592460118             0.8407539882
## 224                High            0.2064348880             0.7935651120
## 231                High            0.7632446817             0.2367553183
## 236                High            0.0866704378             0.9133295622
## 237                High            0.0136419825             0.9863580175
## 240                High            0.2146524994             0.7853475006
## 243                High            0.0436230271             0.9563769729
## 248                High            0.1108113031             0.8891886969
## 251                High            0.6961143000             0.3038857000
## 256                High            0.3070340842             0.6929659158
## 258                High            0.1230410001             0.8769589999
## 262                High            0.5395740091             0.4604259909
## 266                High            0.4571965619             0.5428034381
## 272                High            0.5720886850             0.4279113150
## 280                High            0.3082654815             0.6917345185
## 283                High            0.4354389857             0.5645610143
## 286                High            0.4391327329             0.5608672671
## 287                High            0.1041477820             0.8958522180
## 289                High            0.1437887272             0.8562112728
## 290                High            0.3282018426             0.6717981574
## 298                High            0.2939848627             0.7060151373
## 305                High            0.2492525888             0.7507474112
## 306                High            0.2398758877             0.7601241123
## 312                High            0.0973381705             0.9026618295
## 320                High            0.3058707588             0.6941292412
## 325                High            0.1771522624             0.8228477376
## 332                High            0.0556061812             0.9443938188
## 333                High            0.4847787185             0.5152212815
## 335                High            0.2939848627             0.7060151373
## 339                High            0.8371427968             0.1628572032
## 346                High            0.2382954634             0.7617045366
## 347                High            0.0327311005             0.9672688995
## 350                High            0.4097345492             0.5902654508
## 353                High            0.2663701600             0.7336298400
## 358                High            0.2182251914             0.7817748086
## 365                High            0.1922987093             0.8077012907
## 367                High            0.2058056954             0.7941943046
## 370                High            0.0651543422             0.9348456578
## 379                High            0.1327838973             0.8672161027
## 386                High            0.5247916818             0.4752083182
## 394                High            0.4697661230             0.5302338770
## 396                High            0.1215635914             0.8784364086
## 400                High            0.0136419825             0.9863580175
## 404                High            0.0493865148             0.9506134852
## 405                High            0.6323780570             0.3676219430
## 413                High            0.0931860048             0.9068139952
## 415                High            0.3133445051             0.6866554949
## 417                High            0.2643581720             0.7356418280
## 418                High            0.3679591512             0.6320408488
## 423                High            0.2714888903             0.7285111097
## 434                High            0.1877329621             0.8122670379
## 437                High            0.3396836347             0.6603163653
## 440                High            0.4302378412             0.5697621588
## 449                High            0.3642274184             0.6357725816
## 450                High            0.2638891235             0.7361108765
## 457                High            0.3642274184             0.6357725816
## 467                High            0.3843344069             0.6156655931
## 469                High            0.3427771468             0.6572228532
## 474                High            0.9448977678             0.0551022322
## 475                High            0.9069691143             0.0930308857
## 485                High            0.1206262001             0.8793737999
## 504                 Low            0.2810994647             0.7189005353
## 511                 Low            0.7713879404             0.2286120596
## 512                 Low            0.3771323094             0.6228676906
## 517                 Low            0.0796952208             0.9203047792
## 519                 Low            0.6631915667             0.3368084333
## 520                 Low            0.0365399927             0.9634600073
## 522                 Low            0.9567591567             0.0432408433
## 527                 Low            0.5472028728             0.4527971272
## 528                 Low            0.4455816601             0.5544183399
## 529                 Low            0.2855557521             0.7144442479
## 537                 Low            0.0818301918             0.9181698082
## 540                 Low            0.9324385799             0.0675614201
## 541                 Low            0.7941384873             0.2058615127
## 547                 Low            0.9225916586             0.0774083414
## 550                 Low            0.7281742163             0.2718257837
## 555                 Low            0.4391059978             0.5608940022
## 564                 Low            0.0179846866             0.9820153134
## 570                 Low            0.3080507810             0.6919492190
## 573                 Low            0.2060658806             0.7939341194
## 575                 Low            0.7941384873             0.2058615127
## 578                 Low            0.2347216890             0.7652783110
## 581                 Low            0.2060658806             0.7939341194
## 585                 Low            0.3173015329             0.6826984671
## 590                 Low            0.6875129427             0.3124870573
## 601                 Low            0.7826540228             0.2173459772
## 602                 Low            0.6143605885             0.3856394115
## 607                 Low            0.4573521726             0.5426478274
## 610                 Low            0.4429636059             0.5570363941
## 618                 Low            0.7158484958             0.2841515042
## 624                 Low            0.3173015329             0.6826984671
## 626                 Low            0.1600193605             0.8399806395
## 627                 Low            0.3215503018             0.6784496982
## 634                 Low            0.6883858742             0.3116141258
## 640                 Low            0.9844093361             0.0155906639
## 642                 Low            0.2100918801             0.7899081199
## 643                 Low            0.8692472243             0.1307527757
## 644                 Low            0.8718338373             0.1281661627
## 645                 Low            0.7731505226             0.2268494774
## 646                 Low            0.9397436656             0.0602563344
## 647                 Low            0.7401939612             0.2598060388
## 652                 Low            0.1584162203             0.8415837797
## 658                 Low            0.5345180440             0.4654819560
## 659                 Low            0.9685770972             0.0314229028
## 660                 Low            0.8148166346             0.1851833654
## 664                 Low            0.3515609710             0.6484390290
## 666                 Low            0.4641158585             0.5358841415
## 667                 Low            0.8356306242             0.1643693758
## 675                 Low            0.8692472243             0.1307527757
## 680                 Low            0.9606395556             0.0393604444
## 681                 Low            0.8406617830             0.1593382170
## 687                 Low            0.9754003752             0.0245996248
## 694                 Low            0.6710783028             0.3289216972
## 697                 Low            0.5964862294             0.4035137706
## 701                 Low            0.5884750688             0.4115249312
## 705                 Low            0.8954203536             0.1045796464
## 707                 Low            0.5851454633             0.4148545367
## 710                 Low            0.8010389126             0.1989610874
## 716                 Low            0.8184036632             0.1815963368
## 719                 Low            0.9428266773             0.0571733227
## 720                 Low            0.9847960378             0.0152039622
## 725                 Low            0.9818803020             0.0181196980
## 727                 Low            0.5884750688             0.4115249312
## 730                 Low            0.5611518183             0.4388481817
## 738                 Low            0.9081879879             0.0918120121
## 745                 Low            0.9170977223             0.0829022777
## 748                 Low            0.8391627813             0.1608372187
## 751                 Low            0.8963099718             0.1036900282
## 756                 Low            0.6480019501             0.3519980499
## 766                 Low            0.6647629476             0.3352370524
## 769                 Low            0.6639046057             0.3360953943
## 783                 Low            0.8863780609             0.1136219391
## 785                 Low            0.9847467994             0.0152532006
## 790                 Low            0.8248336297             0.1751663703
## 793                 Low            0.8863780609             0.1136219391
## 795                 Low            0.9897095855             0.0102904145
## 796                 Low            0.9850516420             0.0149483580
## 797                 Low            0.4811591379             0.5188408621
## 801                 Low            0.7986667815             0.2013332185
## 811                 Low            0.5989448209             0.4010551791
## 812                 Low            0.8865111374             0.1134888626
## 815                 Low            0.9820707051             0.0179292949
## 816                 Low            0.8050772583             0.1949227417
## 817                 Low            0.9939985174             0.0060014826
## 824                 Low            0.8901108481             0.1098891519
## 825                 Low            0.8901108481             0.1098891519
## 826                 Low            0.8901108481             0.1098891519
## 830                 Low            0.9863085584             0.0136914416
## 837                 Low            0.9927031149             0.0072968851
## 838                 Low            0.8050772583             0.1949227417
## 844                 Low            0.9899820408             0.0100179592
## 845                 Low            0.9980207841             0.0019792159
## 847                 Low            0.9347533178             0.0652466822
## 850                 Low            0.9018688685             0.0981311315
## 852                 Low            0.9155921637             0.0844078363
## 853                 Low            0.9155921637             0.0844078363
## 861                 Low            0.9288460568             0.0711539432
## 868                 Low            0.9983178675             0.0016821325
## 874                 Low            0.9409936465             0.0590063535
## 879                High            0.0743114149             0.9256885851
## 895                High            0.0058470295             0.9941529705
## 899                High            0.0007541913             0.9992458087
## 903                High            0.0058470295             0.9941529705
## 917                High            0.0541474152             0.9458525848
## 927                High            0.0127138418             0.9872861582
## 929                High            0.0762280843             0.9237719157
## 931                High            0.0127138418             0.9872861582
## 933                High            0.3431573442             0.6568426558
## 944                High            0.0283606163             0.9716393837
## 947                High            0.0246356037             0.9753643963
## 949                High            0.1676976167             0.8323023833
## 953                High            0.0325221723             0.9674778277
## 958                High            0.5930878020             0.4069121980
## 961                High            0.0095523084             0.9904476916
## 963                High            0.0969115592             0.9030884408
## 964                High            0.0436230271             0.9563769729
## 973                High            0.0478986779             0.9521013221
## 976                High            0.0179846866             0.9820153134
## 977                High            0.1108113031             0.8891886969
## 980                High            0.3323177208             0.6676822792
## 983                High            0.6441963861             0.3558036139
## 984                High            0.1108113031             0.8891886969
## 986                High            0.0973381705             0.9026618295
## 989                High            0.1679011559             0.8320988441
## 991                High            0.0218893139             0.9781106861
## 996                High            0.0176152778             0.9823847222
## 997                High            0.4484937124             0.5515062876
## 999                High            0.0354283423             0.9645716577
## 1000               High            0.0598930376             0.9401069624
## 1003               High            0.0136419825             0.9863580175
## 1008               High            0.0805914802             0.9194085198
## 1009               High            0.4354389857             0.5645610143
## 1014               High            0.0282625766             0.9717374234
## 1015               High            0.4604541718             0.5395458282
## 1040               High            0.1834826180             0.8165173820
## 1042               High            0.3502196618             0.6497803382
## 1043               High            0.7301353926             0.2698646074
## 1050               High            0.1128584616             0.8871415384
## 1052               High            0.1326642328             0.8673357672
## 1056               High            0.4214253198             0.5785746802
## 1070               High            0.5016341979             0.4983658021
## 1073               High            0.5016785964             0.4983214036
## 1074               High            0.1240863502             0.8759136498
## 1079               High            0.4087109291             0.5912890709
## 1080               High            0.6775015397             0.3224984603
## 1085               High            0.1017757206             0.8982242794
## 1087               High            0.5255391386             0.4744608614
## 1096               High            0.8643165720             0.1356834280
## 1099               High            0.4099125986             0.5900874014
## 1100               High            0.7154999827             0.2845000173
## 1102               High            0.0887503601             0.9112496399
## 1107                Low            0.6788519384             0.3211480616
## 1109                Low            0.6292550701             0.3707449299
## 1114                Low            0.7123656715             0.2876343285
## 1118                Low            0.2889459081             0.7110540919
## 1123                Low            0.4860246521             0.5139753479
## 1132                Low            0.9313431964             0.0686568036
## 1134                Low            0.5204468329             0.4795531671
## 1137                Low            0.3215503018             0.6784496982
## 1154                Low            0.3215503018             0.6784496982
## 1155                Low            0.4988737937             0.5011262063
## 1157                Low            0.6870639639             0.3129360361
## 1162                Low            0.8692472243             0.1307527757
## 1164                Low            0.2100918801             0.7899081199
## 1171                Low            0.8963099718             0.1036900282
## 1172                Low            0.5491594648             0.4508405352
## 1175                Low            0.5561457281             0.4438542719
## 1177                Low            0.7164966875             0.2835033125
## 1179                Low            0.9138172674             0.0861827326
## 1183                Low            0.4123464196             0.5876535804
## 1185                Low            0.9073388166             0.0926611834
## 1189                Low            0.7755333295             0.2244666705
## 1211                Low            0.9600742008             0.0399257992
## 1218                Low            0.9898292244             0.0101707756
## 1224                Low            0.4045598255             0.5954401745
## 1225                Low            0.5884750688             0.4115249312
## 1227                Low            0.8791559614             0.1208440386
## 1232                Low            0.9872441389             0.0127558611
## 1235                Low            0.6380743568             0.3619256432
## 1238                Low            0.6179361327             0.3820638673
## 1240                Low            0.8199182266             0.1800817734
## 1241                Low            0.9600742008             0.0399257992
## 1248                Low            0.9863085584             0.0136914416
## 1258                Low            0.8050772583             0.1949227417
## 1261                Low            0.9027349430             0.0972650570
## 1263                Low            0.8901108481             0.1098891519
## 1269                Low            0.9568830828             0.0431169172
## 1270                Low            0.9882558677             0.0117441323
## 1271                Low            0.9018688685             0.0981311315
## 1272                Low            0.9155921637             0.0844078363
## 1280                Low            0.9409936465             0.0590063535
## 1286                Low            0.9990096848             0.0009903152
## 1287                Low            0.9990849676             0.0009150324
## 1289                Low            0.9599001175             0.0400998825
## 1290                Low            0.9155921637             0.0844078363
## 1291               High            0.1594073766             0.8405926234
## 1294               High            0.8044109182             0.1955890818
## 1305                Low            0.8477696040             0.1522303960
## 1308               High            0.6914165831             0.3085834169

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_LOG10T_Test_ROC <- roc(response = LR_LOG10T_Test$LR_LOG10T_Observed,
             predictor = LR_LOG10T_Test$LR_LOG10T_Predicted.High,
             levels = rev(levels(LR_LOG10T_Test$LR_LOG10T_Observed)))

(LR_LOG10T_Test_ROCCurveAUC <- auc(LR_LOG10T_Test_ROC)[1])

## [1] 0.8988237

1.5.7 Logistic Regression With Natural Logarithm Transformation (LR_LNT)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Natural Logarithm Transformation applies an algebraic transformation by substituting the variable by its natural logarithm function-applied form using a base equal to the constant e (2.7182818). As the logarithm of zero or any negative number is undefined, a constant must be added to move the minimum value of the distribution preferably to 1.

[A] The logistic regression model from the stats package was implemented through the caret package with natural logarithm transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.89192

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89882

Code Chunk | Output

##################################
# Applying numerical adjustments
# to eliminate zero values and
# natural logarithm function
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  PMA_PreModelling_Train[,i] <- PMA_PreModelling_Train[,i] + 1
  PMA_PreModelling_Train[,i] <- log(PMA_PreModelling_Train[,i])
}

for (i in 1:(ncol(PMA_PreModelling_Test)-1)){
  PMA_PreModelling_Test[,i] <- PMA_PreModelling_Test[,i] + 1
  PMA_PreModelling_Test[,i] <- log(PMA_PreModelling_Test[,i])
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
PMA_PreModelling_Train_LR_LNT <- PMA_PreModelling_Train_LR
PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_LNT[,sapply(PMA_PreModelling_Train_LR_LNT, is.numeric)],
            y = PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_LNT Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_LNT[,sapply(PMA_PreModelling_Train_LR_LNT, is.numeric)], 
            y = PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_LNT Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_LNT_Tune <- train(x = PMA_PreModelling_Train_LR_LNT[,!names(PMA_PreModelling_Train_LR_LNT) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_LNT$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_LNT_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC       Sens       Spec     
##   0.891921  0.7492802  0.8245646

LR_LNT_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##           11.4167             1.2717             1.4241            -6.1614  
##         NumCarbon  
##            0.4056  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 767.4     AIC: 777.4

LR_LNT_Tune$results

##   parameter      ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.891921 0.7492802 0.8245646 0.03393011 0.06711081 0.06719134

(LR_LNT_Train_ROCCurveAUC <- LR_LNT_Tune$results$ROC)

## [1] 0.891921

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_LNT_VarImp <- varImp(LR_LNT_Tune, scale = TRUE)
plot(LR_LNT_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_LNT <- PMA_PreModelling_Test_LR
PMA_PreModelling_Test_LR_LNT$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_LNT[,sapply(PMA_PreModelling_Test_LR_LNT, is.numeric)],
            y = PMA_PreModelling_Test_LR_LNT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_LNT Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_LNT[,sapply(PMA_PreModelling_Test_LR_LNT, is.numeric)], 
            y = PMA_PreModelling_Test_LR_LNT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_LNT Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_LNT$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_LNT_Test <- data.frame(LR_LNT_Observed = PMA_PreModelling_Test_LR_LNT$Log_Solubility_Class,
                      LR_LNT_Predicted = predict(LR_LNT_Tune,
                      PMA_PreModelling_Test_LR_LNT[,!names(PMA_PreModelling_Test_LR_LNT) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_LNT_Test

##      LR_LNT_Observed LR_LNT_Predicted.Low LR_LNT_Predicted.High
## 20              High         0.0114244517          0.9885755483
## 21              High         0.0080310909          0.9919689091
## 23              High         0.0354283423          0.9645716577
## 25              High         0.0113319188          0.9886680812
## 28              High         0.0603784377          0.9396215623
## 31              High         0.0126421937          0.9873578063
## 32              High         0.0204088121          0.9795911879
## 33              High         0.0401610347          0.9598389653
## 34              High         0.0401610347          0.9598389653
## 37              High         0.3543519894          0.6456480106
## 38              High         0.3543519894          0.6456480106
## 42              High         0.4516439293          0.5483560707
## 49              High         0.1326709535          0.8673290465
## 54              High         0.0127138418          0.9872861582
## 55              High         0.0171425230          0.9828574770
## 58              High         0.4508886389          0.5491113611
## 60              High         0.0458725899          0.9541274101
## 61              High         0.0401610347          0.9598389653
## 65              High         0.0422682755          0.9577317245
## 69              High         0.5360931241          0.4639068759
## 73              High         0.0313344498          0.9686655502
## 86              High         0.0379168490          0.9620831510
## 90              High         0.0246356037          0.9753643963
## 91              High         0.0112519351          0.9887480649
## 93              High         0.0246356037          0.9753643963
## 96              High         0.0112519351          0.9887480649
## 98              High         0.0281163829          0.9718836171
## 100             High         0.0455991395          0.9544008605
## 104             High         0.8596843413          0.1403156587
## 112             High         0.1899203318          0.8100796682
## 115             High         0.8830092589          0.1169907411
## 119             High         0.1145296081          0.8854703919
## 128             High         0.1145296081          0.8854703919
## 130             High         0.0186594977          0.9813405023
## 139             High         0.0186594977          0.9813405023
## 143             High         0.0369097104          0.9630902896
## 145             High         0.0632876797          0.9367123203
## 146             High         0.1145296081          0.8854703919
## 149             High         0.1330597875          0.8669402125
## 150             High         0.0881581152          0.9118418848
## 152             High         0.0401610347          0.9598389653
## 157             High         0.5491594648          0.4508405352
## 161             High         0.1818965999          0.8181034001
## 162             High         0.0114108307          0.9885891693
## 166             High         0.5306213331          0.4693786669
## 167             High         0.1130629002          0.8869370998
## 173             High         0.0762280843          0.9237719157
## 176             High         0.1145296081          0.8854703919
## 182             High         0.0040896523          0.9959103477
## 187             High         0.0411811702          0.9588188298
## 190             High         0.0232842897          0.9767157103
## 194             High         0.0175444271          0.9824555729
## 195             High         0.1108113031          0.8891886969
## 201             High         0.0436230271          0.9563769729
## 207             High         0.0716468222          0.9283531778
## 208             High         0.5395740091          0.4604259909
## 215             High         0.0126421937          0.9873578063
## 222             High         0.1592460118          0.8407539882
## 224             High         0.2064348880          0.7935651120
## 231             High         0.7632446817          0.2367553183
## 236             High         0.0866704378          0.9133295622
## 237             High         0.0136419825          0.9863580175
## 240             High         0.2146524994          0.7853475006
## 243             High         0.0436230271          0.9563769729
## 248             High         0.1108113031          0.8891886969
## 251             High         0.6961143000          0.3038857000
## 256             High         0.3070340842          0.6929659158
## 258             High         0.1230410001          0.8769589999
## 262             High         0.5395740091          0.4604259909
## 266             High         0.4571965619          0.5428034381
## 272             High         0.5720886850          0.4279113150
## 280             High         0.3082654815          0.6917345185
## 283             High         0.4354389857          0.5645610143
## 286             High         0.4391327329          0.5608672671
## 287             High         0.1041477820          0.8958522180
## 289             High         0.1437887272          0.8562112728
## 290             High         0.3282018426          0.6717981574
## 298             High         0.2939848627          0.7060151373
## 305             High         0.2492525888          0.7507474112
## 306             High         0.2398758877          0.7601241123
## 312             High         0.0973381705          0.9026618295
## 320             High         0.3058707588          0.6941292412
## 325             High         0.1771522624          0.8228477376
## 332             High         0.0556061812          0.9443938188
## 333             High         0.4847787185          0.5152212815
## 335             High         0.2939848627          0.7060151373
## 339             High         0.8371427968          0.1628572032
## 346             High         0.2382954634          0.7617045366
## 347             High         0.0327311005          0.9672688995
## 350             High         0.4097345492          0.5902654508
## 353             High         0.2663701600          0.7336298400
## 358             High         0.2182251914          0.7817748086
## 365             High         0.1922987093          0.8077012907
## 367             High         0.2058056954          0.7941943046
## 370             High         0.0651543422          0.9348456578
## 379             High         0.1327838973          0.8672161027
## 386             High         0.5247916818          0.4752083182
## 394             High         0.4697661230          0.5302338770
## 396             High         0.1215635914          0.8784364086
## 400             High         0.0136419825          0.9863580175
## 404             High         0.0493865148          0.9506134852
## 405             High         0.6323780570          0.3676219430
## 413             High         0.0931860048          0.9068139952
## 415             High         0.3133445051          0.6866554949
## 417             High         0.2643581720          0.7356418280
## 418             High         0.3679591512          0.6320408488
## 423             High         0.2714888903          0.7285111097
## 434             High         0.1877329621          0.8122670379
## 437             High         0.3396836347          0.6603163653
## 440             High         0.4302378412          0.5697621588
## 449             High         0.3642274184          0.6357725816
## 450             High         0.2638891235          0.7361108765
## 457             High         0.3642274184          0.6357725816
## 467             High         0.3843344069          0.6156655931
## 469             High         0.3427771468          0.6572228532
## 474             High         0.9448977678          0.0551022322
## 475             High         0.9069691143          0.0930308857
## 485             High         0.1206262001          0.8793737999
## 504              Low         0.2810994647          0.7189005353
## 511              Low         0.7713879404          0.2286120596
## 512              Low         0.3771323094          0.6228676906
## 517              Low         0.0796952208          0.9203047792
## 519              Low         0.6631915667          0.3368084333
## 520              Low         0.0365399927          0.9634600073
## 522              Low         0.9567591567          0.0432408433
## 527              Low         0.5472028728          0.4527971272
## 528              Low         0.4455816601          0.5544183399
## 529              Low         0.2855557521          0.7144442479
## 537              Low         0.0818301918          0.9181698082
## 540              Low         0.9324385799          0.0675614201
## 541              Low         0.7941384873          0.2058615127
## 547              Low         0.9225916586          0.0774083414
## 550              Low         0.7281742163          0.2718257837
## 555              Low         0.4391059978          0.5608940022
## 564              Low         0.0179846866          0.9820153134
## 570              Low         0.3080507810          0.6919492190
## 573              Low         0.2060658806          0.7939341194
## 575              Low         0.7941384873          0.2058615127
## 578              Low         0.2347216890          0.7652783110
## 581              Low         0.2060658806          0.7939341194
## 585              Low         0.3173015329          0.6826984671
## 590              Low         0.6875129427          0.3124870573
## 601              Low         0.7826540228          0.2173459772
## 602              Low         0.6143605885          0.3856394115
## 607              Low         0.4573521726          0.5426478274
## 610              Low         0.4429636059          0.5570363941
## 618              Low         0.7158484958          0.2841515042
## 624              Low         0.3173015329          0.6826984671
## 626              Low         0.1600193605          0.8399806395
## 627              Low         0.3215503018          0.6784496982
## 634              Low         0.6883858742          0.3116141258
## 640              Low         0.9844093361          0.0155906639
## 642              Low         0.2100918801          0.7899081199
## 643              Low         0.8692472243          0.1307527757
## 644              Low         0.8718338373          0.1281661627
## 645              Low         0.7731505226          0.2268494774
## 646              Low         0.9397436656          0.0602563344
## 647              Low         0.7401939612          0.2598060388
## 652              Low         0.1584162203          0.8415837797
## 658              Low         0.5345180440          0.4654819560
## 659              Low         0.9685770972          0.0314229028
## 660              Low         0.8148166346          0.1851833654
## 664              Low         0.3515609710          0.6484390290
## 666              Low         0.4641158585          0.5358841415
## 667              Low         0.8356306242          0.1643693758
## 675              Low         0.8692472243          0.1307527757
## 680              Low         0.9606395556          0.0393604444
## 681              Low         0.8406617830          0.1593382170
## 687              Low         0.9754003752          0.0245996248
## 694              Low         0.6710783028          0.3289216972
## 697              Low         0.5964862294          0.4035137706
## 701              Low         0.5884750688          0.4115249312
## 705              Low         0.8954203536          0.1045796464
## 707              Low         0.5851454633          0.4148545367
## 710              Low         0.8010389126          0.1989610874
## 716              Low         0.8184036632          0.1815963368
## 719              Low         0.9428266773          0.0571733227
## 720              Low         0.9847960378          0.0152039622
## 725              Low         0.9818803020          0.0181196980
## 727              Low         0.5884750688          0.4115249312
## 730              Low         0.5611518183          0.4388481817
## 738              Low         0.9081879879          0.0918120121
## 745              Low         0.9170977223          0.0829022777
## 748              Low         0.8391627813          0.1608372187
## 751              Low         0.8963099718          0.1036900282
## 756              Low         0.6480019501          0.3519980499
## 766              Low         0.6647629476          0.3352370524
## 769              Low         0.6639046057          0.3360953943
## 783              Low         0.8863780609          0.1136219391
## 785              Low         0.9847467994          0.0152532006
## 790              Low         0.8248336297          0.1751663703
## 793              Low         0.8863780609          0.1136219391
## 795              Low         0.9897095855          0.0102904145
## 796              Low         0.9850516420          0.0149483580
## 797              Low         0.4811591379          0.5188408621
## 801              Low         0.7986667815          0.2013332185
## 811              Low         0.5989448209          0.4010551791
## 812              Low         0.8865111374          0.1134888626
## 815              Low         0.9820707051          0.0179292949
## 816              Low         0.8050772583          0.1949227417
## 817              Low         0.9939985174          0.0060014826
## 824              Low         0.8901108481          0.1098891519
## 825              Low         0.8901108481          0.1098891519
## 826              Low         0.8901108481          0.1098891519
## 830              Low         0.9863085584          0.0136914416
## 837              Low         0.9927031149          0.0072968851
## 838              Low         0.8050772583          0.1949227417
## 844              Low         0.9899820408          0.0100179592
## 845              Low         0.9980207841          0.0019792159
## 847              Low         0.9347533178          0.0652466822
## 850              Low         0.9018688685          0.0981311315
## 852              Low         0.9155921637          0.0844078363
## 853              Low         0.9155921637          0.0844078363
## 861              Low         0.9288460568          0.0711539432
## 868              Low         0.9983178675          0.0016821325
## 874              Low         0.9409936465          0.0590063535
## 879             High         0.0743114149          0.9256885851
## 895             High         0.0058470295          0.9941529705
## 899             High         0.0007541913          0.9992458087
## 903             High         0.0058470295          0.9941529705
## 917             High         0.0541474152          0.9458525848
## 927             High         0.0127138418          0.9872861582
## 929             High         0.0762280843          0.9237719157
## 931             High         0.0127138418          0.9872861582
## 933             High         0.3431573442          0.6568426558
## 944             High         0.0283606163          0.9716393837
## 947             High         0.0246356037          0.9753643963
## 949             High         0.1676976167          0.8323023833
## 953             High         0.0325221723          0.9674778277
## 958             High         0.5930878020          0.4069121980
## 961             High         0.0095523084          0.9904476916
## 963             High         0.0969115592          0.9030884408
## 964             High         0.0436230271          0.9563769729
## 973             High         0.0478986779          0.9521013221
## 976             High         0.0179846866          0.9820153134
## 977             High         0.1108113031          0.8891886969
## 980             High         0.3323177208          0.6676822792
## 983             High         0.6441963861          0.3558036139
## 984             High         0.1108113031          0.8891886969
## 986             High         0.0973381705          0.9026618295
## 989             High         0.1679011559          0.8320988441
## 991             High         0.0218893139          0.9781106861
## 996             High         0.0176152778          0.9823847222
## 997             High         0.4484937124          0.5515062876
## 999             High         0.0354283423          0.9645716577
## 1000            High         0.0598930376          0.9401069624
## 1003            High         0.0136419825          0.9863580175
## 1008            High         0.0805914802          0.9194085198
## 1009            High         0.4354389857          0.5645610143
## 1014            High         0.0282625766          0.9717374234
## 1015            High         0.4604541718          0.5395458282
## 1040            High         0.1834826180          0.8165173820
## 1042            High         0.3502196618          0.6497803382
## 1043            High         0.7301353926          0.2698646074
## 1050            High         0.1128584616          0.8871415384
## 1052            High         0.1326642328          0.8673357672
## 1056            High         0.4214253198          0.5785746802
## 1070            High         0.5016341979          0.4983658021
## 1073            High         0.5016785964          0.4983214036
## 1074            High         0.1240863502          0.8759136498
## 1079            High         0.4087109291          0.5912890709
## 1080            High         0.6775015397          0.3224984603
## 1085            High         0.1017757206          0.8982242794
## 1087            High         0.5255391386          0.4744608614
## 1096            High         0.8643165720          0.1356834280
## 1099            High         0.4099125986          0.5900874014
## 1100            High         0.7154999827          0.2845000173
## 1102            High         0.0887503601          0.9112496399
## 1107             Low         0.6788519384          0.3211480616
## 1109             Low         0.6292550701          0.3707449299
## 1114             Low         0.7123656715          0.2876343285
## 1118             Low         0.2889459081          0.7110540919
## 1123             Low         0.4860246521          0.5139753479
## 1132             Low         0.9313431964          0.0686568036
## 1134             Low         0.5204468329          0.4795531671
## 1137             Low         0.3215503018          0.6784496982
## 1154             Low         0.3215503018          0.6784496982
## 1155             Low         0.4988737937          0.5011262063
## 1157             Low         0.6870639639          0.3129360361
## 1162             Low         0.8692472243          0.1307527757
## 1164             Low         0.2100918801          0.7899081199
## 1171             Low         0.8963099718          0.1036900282
## 1172             Low         0.5491594648          0.4508405352
## 1175             Low         0.5561457281          0.4438542719
## 1177             Low         0.7164966875          0.2835033125
## 1179             Low         0.9138172674          0.0861827326
## 1183             Low         0.4123464196          0.5876535804
## 1185             Low         0.9073388166          0.0926611834
## 1189             Low         0.7755333295          0.2244666705
## 1211             Low         0.9600742008          0.0399257992
## 1218             Low         0.9898292244          0.0101707756
## 1224             Low         0.4045598255          0.5954401745
## 1225             Low         0.5884750688          0.4115249312
## 1227             Low         0.8791559614          0.1208440386
## 1232             Low         0.9872441389          0.0127558611
## 1235             Low         0.6380743568          0.3619256432
## 1238             Low         0.6179361327          0.3820638673
## 1240             Low         0.8199182266          0.1800817734
## 1241             Low         0.9600742008          0.0399257992
## 1248             Low         0.9863085584          0.0136914416
## 1258             Low         0.8050772583          0.1949227417
## 1261             Low         0.9027349430          0.0972650570
## 1263             Low         0.8901108481          0.1098891519
## 1269             Low         0.9568830828          0.0431169172
## 1270             Low         0.9882558677          0.0117441323
## 1271             Low         0.9018688685          0.0981311315
## 1272             Low         0.9155921637          0.0844078363
## 1280             Low         0.9409936465          0.0590063535
## 1286             Low         0.9990096848          0.0009903152
## 1287             Low         0.9990849676          0.0009150324
## 1289             Low         0.9599001175          0.0400998825
## 1290             Low         0.9155921637          0.0844078363
## 1291            High         0.1594073766          0.8405926234
## 1294            High         0.8044109182          0.1955890818
## 1305             Low         0.8477696040          0.1522303960
## 1308            High         0.6914165831          0.3085834169

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_LNT_Test_ROC <- roc(response = LR_LNT_Test$LR_LNT_Observed,
             predictor = LR_LNT_Test$LR_LNT_Predicted.High,
             levels = rev(levels(LR_LNT_Test$LR_LNT_Observed)))

(LR_LNT_Test_ROCCurveAUC <- auc(LR_LNT_Test_ROC)[1])

## [1] 0.8988237

1.5.8 Logistic Regression With Square Root Transformation (LR_SRT)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Square Root Transformation applies an algebraic transformation by substituting the variable by its square root function-applied form. While the function can be applied to zero values, square root of any negative number is undefined, a constant must be added to move the minimum value of the distribution preferably to 0.

[A] The logistic regression model from the stats package was implemented through the caret package with square root transformation applied to treat data skewness but no any treatment applied for data outliers.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.88441

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.89882

Code Chunk | Output

##################################
# Applying square root function
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  PMA_PreModelling_Train[,i] <- PMA_PreModelling_Train[,i] + 1
  PMA_PreModelling_Train[,i] <- sqrt(PMA_PreModelling_Train[,i])
}

for (i in 1:(ncol(PMA_PreModelling_Test)-1)){
  PMA_PreModelling_Test[,i] <- PMA_PreModelling_Test[,i] + 1
  PMA_PreModelling_Test[,i] <- sqrt(PMA_PreModelling_Test[,i])
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
PMA_PreModelling_Train_LR_SRT <- PMA_PreModelling_Train_LR
PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the train set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_SRT[,sapply(PMA_PreModelling_Train_LR_SRT, is.numeric)],
            y = PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_SRT Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_SRT[,sapply(PMA_PreModelling_Train_LR_SRT, is.numeric)], 
            y = PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_SRT Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_SRT_Tune <- train(x = PMA_PreModelling_Train_LR_SRT[,!names(PMA_PreModelling_Train_LR_SRT) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_SRT$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_SRT_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec    
##   0.8844149  0.7163344  0.830225

LR_SRT_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##            6.2882             2.7830             0.5132            -2.7168  
##         NumCarbon  
##           -0.3034  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 800.3     AIC: 810.3

LR_SRT_Tune$results

##   parameter       ROC      Sens     Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8844149 0.7163344 0.830225 0.03385158 0.07539679 0.06335689

(LR_SRT_Train_ROCCurveAUC <- LR_SRT_Tune$results$ROC)

## [1] 0.8844149

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_SRT_VarImp <- varImp(LR_SRT_Tune, scale = TRUE)
plot(LR_SRT_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
PMA_PreModelling_Test_LR_SRT <- PMA_PreModelling_Test_LR
PMA_PreModelling_Test_LR_SRT$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Treating data outliers
# for the test set
##################################
# No actions applied

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_SRT[,sapply(PMA_PreModelling_Test_LR_SRT, is.numeric)],
            y = PMA_PreModelling_Test_LR_SRT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_SRT Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_SRT[,sapply(PMA_PreModelling_Test_LR_SRT, is.numeric)], 
            y = PMA_PreModelling_Test_LR_SRT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_SRT Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_SRT$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_SRT_Test <- data.frame(LR_SRT_Observed = PMA_PreModelling_Test_LR_SRT$Log_Solubility_Class,
                      LR_SRT_Predicted = predict(LR_SRT_Tune,
                      PMA_PreModelling_Test_LR_SRT[,!names(PMA_PreModelling_Test_LR_SRT) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_SRT_Test

##      LR_SRT_Observed LR_SRT_Predicted.Low LR_SRT_Predicted.High
## 20              High         0.0171776680           0.982822332
## 21              High         0.0094733300           0.990526670
## 23              High         0.0429512262           0.957048774
## 25              High         0.0166962209           0.983303779
## 28              High         0.0615130927           0.938486907
## 31              High         0.0346712284           0.965328772
## 32              High         0.0572263673           0.942773633
## 33              High         0.0870345720           0.912965428
## 34              High         0.0870345720           0.912965428
## 37              High         0.3518170970           0.648182903
## 38              High         0.3518170970           0.648182903
## 42              High         0.5218061033           0.478193897
## 49              High         0.1799027090           0.820097291
## 54              High         0.0328468786           0.967153121
## 55              High         0.0151348891           0.984865111
## 58              High         0.4317935721           0.568206428
## 60              High         0.0847834160           0.915216584
## 61              High         0.0870345720           0.912965428
## 65              High         0.0573837578           0.942616242
## 69              High         0.5187323178           0.481267682
## 73              High         0.0088225257           0.991177474
## 86              High         0.0731684575           0.926831543
## 90              High         0.0538625246           0.946137475
## 91              High         0.0064373829           0.993562617
## 93              High         0.0538625246           0.946137475
## 96              High         0.0064373829           0.993562617
## 98              High         0.0598648467           0.940135153
## 100             High         0.0719361787           0.928063821
## 104             High         0.9099158472           0.090084153
## 112             High         0.2285023414           0.771497659
## 115             High         0.8750053683           0.124994632
## 119             High         0.1622573557           0.837742644
## 128             High         0.1622573557           0.837742644
## 130             High         0.0161060102           0.983893990
## 139             High         0.0161060102           0.983893990
## 143             High         0.0417491435           0.958250856
## 145             High         0.1137434063           0.886256594
## 146             High         0.1622573557           0.837742644
## 149             High         0.1593165890           0.840683411
## 150             High         0.1398798250           0.860120175
## 152             High         0.0870345720           0.912965428
## 157             High         0.5055646637           0.494435336
## 161             High         0.2398674752           0.760132525
## 162             High         0.0066302108           0.993369789
## 166             High         0.5237831608           0.476216839
## 167             High         0.1651313064           0.834868694
## 173             High         0.1259453369           0.874054663
## 176             High         0.1622573557           0.837742644
## 182             High         0.0222998583           0.977700142
## 187             High         0.0370880559           0.962911944
## 190             High         0.0134601255           0.986539874
## 194             High         0.0291152071           0.970884793
## 195             High         0.1750006699           0.824999330
## 201             High         0.0832309264           0.916769074
## 207             High         0.1222726588           0.877727341
## 208             High         0.4877656097           0.512234390
## 215             High         0.0346712284           0.965328772
## 222             High         0.2353885107           0.764611489
## 224             High         0.1922505225           0.807749477
## 231             High         0.7189699421           0.281030058
## 236             High         0.1056715454           0.894328455
## 237             High         0.0456519036           0.954348096
## 240             High         0.2128438968           0.787156103
## 243             High         0.0832309264           0.916769074
## 248             High         0.1750006699           0.824999330
## 251             High         0.7704009245           0.229599076
## 256             High         0.3841407732           0.615859227
## 258             High         0.1701526065           0.829847393
## 262             High         0.4877656097           0.512234390
## 266             High         0.4606389841           0.539361016
## 272             High         0.5616128480           0.438387152
## 280             High         0.3111471891           0.688852811
## 283             High         0.4319894761           0.568010524
## 286             High         0.4462380183           0.553761982
## 287             High         0.1669680937           0.833031906
## 289             High         0.1191373826           0.880862617
## 290             High         0.3312408721           0.668759128
## 298             High         0.3172571574           0.682742843
## 305             High         0.2944486733           0.705551327
## 306             High         0.1943816153           0.805618385
## 312             High         0.0989000206           0.901099979
## 320             High         0.3131330745           0.686866926
## 325             High         0.1824338561           0.817566144
## 332             High         0.0623662697           0.937633730
## 333             High         0.4443815794           0.555618421
## 335             High         0.3172571574           0.682742843
## 339             High         0.7526842928           0.247315707
## 346             High         0.3267166104           0.673283390
## 347             High         0.0533733809           0.946626619
## 350             High         0.3802631772           0.619736823
## 353             High         0.3335196993           0.666480301
## 358             High         0.3045483863           0.695451614
## 365             High         0.2268214855           0.773178514
## 367             High         0.1951627467           0.804837253
## 370             High         0.0351733764           0.964826624
## 379             High         0.1045430079           0.895456992
## 386             High         0.3769454782           0.623054522
## 394             High         0.5261865734           0.473813427
## 396             High         0.1296532429           0.870346757
## 400             High         0.0456519036           0.954348096
## 404             High         0.0482692739           0.951730726
## 405             High         0.5987312318           0.401268768
## 413             High         0.1269081773           0.873091823
## 415             High         0.3684495301           0.631550470
## 417             High         0.2322587979           0.767741202
## 418             High         0.4327304582           0.567269542
## 423             High         0.2911428364           0.708857164
## 434             High         0.2665859841           0.733414016
## 437             High         0.2526126299           0.747387370
## 440             High         0.3719172425           0.628082757
## 449             High         0.3790499525           0.620950048
## 450             High         0.2818076021           0.718192398
## 457             High         0.3790499525           0.620950048
## 467             High         0.3100685455           0.689931454
## 469             High         0.1599358543           0.840064146
## 474             High         0.9353877385           0.064612262
## 475             High         0.9012220999           0.098777900
## 485             High         0.1079410717           0.892058928
## 504              Low         0.1987091431           0.801290857
## 511              Low         0.7040002637           0.295999736
## 512              Low         0.4183441677           0.581655832
## 517              Low         0.0403857448           0.959614255
## 519              Low         0.7193762711           0.280623729
## 520              Low         0.0818263209           0.918173679
## 522              Low         0.9390148916           0.060985108
## 527              Low         0.6225909545           0.377409046
## 528              Low         0.3219492221           0.678050778
## 529              Low         0.3079150489           0.692084951
## 537              Low         0.1326610454           0.867338955
## 540              Low         0.9212090082           0.078790992
## 541              Low         0.5642976565           0.435702343
## 547              Low         0.9188139959           0.081186004
## 550              Low         0.5753601845           0.424639815
## 555              Low         0.5010132937           0.498986706
## 564              Low         0.0412981629           0.958701837
## 570              Low         0.3533886536           0.646611346
## 573              Low         0.2380917144           0.761908286
## 575              Low         0.5642976565           0.435702343
## 578              Low         0.2113833669           0.788616633
## 581              Low         0.2380917144           0.761908286
## 585              Low         0.3217351998           0.678264800
## 590              Low         0.7367369732           0.263263027
## 601              Low         0.8572092918           0.142790708
## 602              Low         0.6397712472           0.360228753
## 607              Low         0.5539816043           0.446018396
## 610              Low         0.5411389615           0.458861038
## 618              Low         0.7487472889           0.251252711
## 624              Low         0.3217351998           0.678264800
## 626              Low         0.2312349798           0.768765020
## 627              Low         0.3353599489           0.664640051
## 634              Low         0.6280533141           0.371946686
## 640              Low         0.9879769391           0.012023061
## 642              Low         0.1928813704           0.807118630
## 643              Low         0.6470732658           0.352926734
## 644              Low         0.8524420814           0.147557919
## 645              Low         0.7576589388           0.242341061
## 646              Low         0.7766825260           0.223317474
## 647              Low         0.7768079181           0.223192082
## 652              Low         0.1992101897           0.800789810
## 658              Low         0.5871468529           0.412853147
## 659              Low         0.8531651594           0.146834841
## 660              Low         0.8805707118           0.119429288
## 664              Low         0.3430474939           0.656952506
## 666              Low         0.4365034540           0.563496546
## 667              Low         0.8629695776           0.137030422
## 675              Low         0.6470732658           0.352926734
## 680              Low         0.9839734149           0.016026585
## 681              Low         0.8925205505           0.107479449
## 687              Low         0.8732490620           0.126750938
## 694              Low         0.7386214676           0.261378532
## 697              Low         0.5743225447           0.425677455
## 701              Low         0.3916949690           0.608305031
## 705              Low         0.9524371969           0.047562803
## 707              Low         0.6657996248           0.334200375
## 710              Low         0.7143156897           0.285684310
## 716              Low         0.8822175146           0.117782485
## 719              Low         0.9406540923           0.059345908
## 720              Low         0.9780703756           0.021929624
## 725              Low         0.9799007961           0.020099204
## 727              Low         0.3916949690           0.608305031
## 730              Low         0.5185168218           0.481483178
## 738              Low         0.8416972687           0.158302731
## 745              Low         0.7186886650           0.281311335
## 748              Low         0.7852185869           0.214781413
## 751              Low         0.9427266890           0.057273311
## 756              Low         0.7000527225           0.299947278
## 766              Low         0.7710747011           0.228925299
## 769              Low         0.5964903375           0.403509663
## 783              Low         0.8502174908           0.149782509
## 785              Low         0.9090787984           0.090921202
## 790              Low         0.8886512925           0.111348707
## 793              Low         0.8502174908           0.149782509
## 795              Low         0.9915592821           0.008440718
## 796              Low         0.9837678710           0.016232129
## 797              Low         0.5608208122           0.439179188
## 801              Low         0.7642200416           0.235779958
## 811              Low         0.5397913514           0.460208649
## 812              Low         0.9363960470           0.063603953
## 815              Low         0.9586825149           0.041317485
## 816              Low         0.7804673310           0.219532669
## 817              Low         0.9566695569           0.043330443
## 824              Low         0.8709603992           0.129039601
## 825              Low         0.8709603992           0.129039601
## 826              Low         0.8709603992           0.129039601
## 830              Low         0.9170635243           0.082936476
## 837              Low         0.9480386701           0.051961330
## 838              Low         0.7804673310           0.219532669
## 844              Low         0.9343242730           0.065675727
## 845              Low         0.9833504460           0.016649554
## 847              Low         0.9315521332           0.068447867
## 850              Low         0.8919647214           0.108035279
## 852              Low         0.9110529484           0.088947052
## 853              Low         0.9110529484           0.088947052
## 861              Low         0.9275671791           0.072432821
## 868              Low         0.9859660498           0.014033950
## 874              Low         0.9416548705           0.058345129
## 879             High         0.1190842356           0.880915764
## 895             High         0.0185473665           0.981452634
## 899             High         0.0002913624           0.999708638
## 903             High         0.0185473665           0.981452634
## 917             High         0.0406554150           0.959344585
## 927             High         0.0328468786           0.967153121
## 929             High         0.1259453369           0.874054663
## 931             High         0.0328468786           0.967153121
## 933             High         0.3687867543           0.631213246
## 944             High         0.0424837446           0.957516255
## 947             High         0.0538625246           0.946137475
## 949             High         0.1819006993           0.818099301
## 953             High         0.0238344707           0.976165529
## 958             High         0.5223971717           0.477602828
## 961             High         0.0171642639           0.982835736
## 963             High         0.1328530189           0.867146981
## 964             High         0.0832309264           0.916769074
## 973             High         0.0517005568           0.948299443
## 976             High         0.0412981629           0.958701837
## 977             High         0.1750006699           0.824999330
## 980             High         0.2633610863           0.736638914
## 983             High         0.6133383461           0.386661654
## 984             High         0.1750006699           0.824999330
## 986             High         0.0989000206           0.901099979
## 989             High         0.2084058286           0.791594171
## 991             High         0.0254199950           0.974580005
## 996             High         0.0528955326           0.947104467
## 997             High         0.4403169291           0.559683071
## 999             High         0.0429512262           0.957048774
## 1000            High         0.0525655462           0.947434454
## 1003            High         0.0456519036           0.954348096
## 1008            High         0.0967426835           0.903257316
## 1009            High         0.4319894761           0.568010524
## 1014            High         0.0429687572           0.957031243
## 1015            High         0.4919848875           0.508015113
## 1040            High         0.1777604857           0.822239514
## 1042            High         0.3526828399           0.647317160
## 1043            High         0.7096112136           0.290388786
## 1050            High         0.1085884359           0.891411564
## 1052            High         0.1792599161           0.820740084
## 1056            High         0.1318850763           0.868114924
## 1070            High         0.5591896277           0.440810372
## 1073            High         0.4657303100           0.534269690
## 1074            High         0.1749873003           0.825012700
## 1079            High         0.4074777876           0.592522212
## 1080            High         0.5848037333           0.415196267
## 1085            High         0.0954182731           0.904581727
## 1087            High         0.6230391933           0.376960807
## 1096            High         0.9185149388           0.081485061
## 1099            High         0.4496281245           0.550371875
## 1100            High         0.6737379747           0.326262025
## 1102            High         0.0589896088           0.941010391
## 1107             Low         0.4722318025           0.527768198
## 1109             Low         0.6900698356           0.309930164
## 1114             Low         0.5024548018           0.497545198
## 1118             Low         0.3450298942           0.654970106
## 1123             Low         0.4711940701           0.528805930
## 1132             Low         0.8749560859           0.125043914
## 1134             Low         0.5948511032           0.405148897
## 1137             Low         0.3353599489           0.664640051
## 1154             Low         0.3353599489           0.664640051
## 1155             Low         0.5731491992           0.426850801
## 1157             Low         0.7851027520           0.214897248
## 1162             Low         0.6470732658           0.352926734
## 1164             Low         0.1928813704           0.807118630
## 1171             Low         0.9427266890           0.057273311
## 1172             Low         0.5055646637           0.494435336
## 1175             Low         0.6405974664           0.359402534
## 1177             Low         0.6626098078           0.337390192
## 1179             Low         0.9406173547           0.059382645
## 1183             Low         0.2979985007           0.702001499
## 1185             Low         0.9489202020           0.051079798
## 1189             Low         0.8642810191           0.135718981
## 1211             Low         0.8242523053           0.175747695
## 1218             Low         0.9938658969           0.006134103
## 1224             Low         0.3956745407           0.604325459
## 1225             Low         0.3916949690           0.608305031
## 1227             Low         0.9249686768           0.075031323
## 1232             Low         0.9889737070           0.011026293
## 1235             Low         0.7256637224           0.274336278
## 1238             Low         0.7264484488           0.273551551
## 1240             Low         0.8721554250           0.127844575
## 1241             Low         0.8242523053           0.175747695
## 1248             Low         0.9170635243           0.082936476
## 1258             Low         0.7804673310           0.219532669
## 1261             Low         0.8974384750           0.102561525
## 1263             Low         0.8709603992           0.129039601
## 1269             Low         0.9605579303           0.039442070
## 1270             Low         0.9897858307           0.010214169
## 1271             Low         0.8919647214           0.108035279
## 1272             Low         0.9110529484           0.088947052
## 1280             Low         0.9416548705           0.058345129
## 1286             Low         0.9913833133           0.008616687
## 1287             Low         0.9920841459           0.007915854
## 1289             Low         0.9625961144           0.037403886
## 1290             Low         0.9110529484           0.088947052
## 1291            High         0.1741556081           0.825844392
## 1294            High         0.7811147653           0.218885235
## 1305             Low         0.9179685932           0.082031407
## 1308            High         0.7657936357           0.234206364

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_SRT_Test_ROC <- roc(response = LR_SRT_Test$LR_SRT_Observed,
             predictor = LR_SRT_Test$LR_SRT_Predicted.High,
             levels = rev(levels(LR_SRT_Test$LR_SRT_Observed)))

(LR_SRT_Test_ROCCurveAUC <- auc(LR_SRT_Test_ROC)[1])

## [1] 0.8946198

1.5.9 Logistic Regression With Outlier Winsorization Treatment (LR_WT)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Winsorization Treatment replaces the extreme values of a data set in order to limit the effect of the outliers on the calculations or the results obtained by using such data. An iteration of such method is an 80% winsorization which replaces the top and bottom 10% data with the values at the 10th and 90th percentiles, respectively. The winsorization process results in a statistical measure of location that is closer to the bulk of the distribution.

[A] The logistic regression model from the stats package was implemented through the caret package with Winsorization transformation applied to treat data outliers but no any treatment applied for data skewness.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.87428

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.88916

Code Chunk | Output

##################################
# Applying winsorization function
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

for (i in 1:(ncol(PMA_PreModelling_Train)-1)){
  Predictor_Percentile90 <- quantile(PMA_PreModelling_Train[,i], 0.90)
  Predictor_Percentile10 <- quantile(PMA_PreModelling_Train[,i], 0.10)
  Predictor_Percentile75 <- quantile(PMA_PreModelling_Train[,i], 0.75)
  Predictor_Percentile25 <- quantile(PMA_PreModelling_Train[,i], 0.25)
  Predictor_IQR <- Predictor_Percentile75-Predictor_Percentile25
  Predictor_Outlier_UCL <- Predictor_Percentile75 + (1.5*Predictor_IQR)
  Predictor_Outlier_LCL <- Predictor_Percentile25 - (1.5*Predictor_IQR)
  PMA_PreModelling_Train[,i] <- ifelse(PMA_PreModelling_Train[,i]>Predictor_Outlier_UCL,Predictor_Percentile90,
                                       ifelse(PMA_PreModelling_Train[,i]<Predictor_Outlier_LCL,Predictor_Percentile10,
                                              PMA_PreModelling_Train[,i]))
  PMA_PreModelling_Test[,i] <- ifelse(PMA_PreModelling_Test[,i]>Predictor_Outlier_UCL,Predictor_Percentile90,
                                       ifelse(PMA_PreModelling_Test[,i]<Predictor_Outlier_LCL,Predictor_Percentile10,
                                              PMA_PreModelling_Test[,i]))
}

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
# No actions applied

##################################
# Treating data outliers
# for the train set
##################################
PMA_PreModelling_Train_LR_WT <- PMA_PreModelling_Train_LR
PMA_PreModelling_Train_LR_WT$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_WT[,sapply(PMA_PreModelling_Train_LR_WT, is.numeric)],
            y = PMA_PreModelling_Train_LR_WT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_WT Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_WT[,sapply(PMA_PreModelling_Train_LR_WT, is.numeric)], 
            y = PMA_PreModelling_Train_LR_WT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_WT Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_WT$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_WT$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_WT_Tune <- train(x = PMA_PreModelling_Train_LR_WT[,!names(PMA_PreModelling_Train_LR_WT) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_WT$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_WT_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8742874  0.7024363  0.8359216

LR_WT_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##           4.21806            1.35203            0.05381           -0.30758  
##         NumCarbon  
##          -0.11897  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 834.5     AIC: 844.5

LR_WT_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8742874 0.7024363 0.8359216 0.03831246 0.06100711 0.06260559

(LR_WT_Train_ROCCurveAUC <- LR_WT_Tune$results$ROC)

## [1] 0.8742874

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_WT_VarImp <- varImp(LR_WT_Tune, scale = TRUE)
plot(LR_WT_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
# No actions applied

##################################
# Treating data outliers
# for the test set
##################################
PMA_PreModelling_Test_LR_WT <- PMA_PreModelling_Test_LR
PMA_PreModelling_Test_LR_WT$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_WT[,sapply(PMA_PreModelling_Test_LR_WT, is.numeric)],
            y = PMA_PreModelling_Test_LR_WT$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_WT Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_WT[,sapply(PMA_PreModelling_Test_LR_WT, is.numeric)], 
            y = PMA_PreModelling_Test_LR_WT$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_WT Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_WT$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_WT_Test <- data.frame(LR_WT_Observed = PMA_PreModelling_Test_LR_WT$Log_Solubility_Class,
                      LR_WT_Predicted = predict(LR_WT_Tune,
                      PMA_PreModelling_Test_LR_WT[,!names(PMA_PreModelling_Test_LR_WT) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_WT_Test

##      LR_WT_Observed LR_WT_Predicted.Low LR_WT_Predicted.High
## 20             High         0.028252639          0.971747361
## 21             High         0.011032277          0.988967723
## 23             High         0.050174667          0.949825333
## 25             High         0.021395292          0.978604708
## 28             High         0.063987262          0.936012738
## 31             High         0.069295112          0.930704888
## 32             High         0.100841511          0.899158489
## 33             High         0.126560560          0.873439440
## 34             High         0.126560560          0.873439440
## 37             High         0.327406425          0.672593575
## 38             High         0.327406425          0.672593575
## 42             High         0.551750285          0.448249715
## 49             High         0.201947838          0.798052162
## 54             High         0.059988529          0.940011471
## 55             High         0.011554148          0.988445852
## 58             High         0.400584308          0.599415692
## 60             High         0.126557798          0.873442202
## 61             High         0.126560560          0.873439440
## 65             High         0.077399182          0.922600818
## 69             High         0.478379139          0.521620861
## 73             High         0.042117210          0.957882790
## 86             High         0.111173578          0.888826422
## 90             High         0.086098117          0.913901883
## 91             High         0.024835727          0.975164273
## 93             High         0.086098117          0.913901883
## 96             High         0.024835727          0.975164273
## 98             High         0.094952344          0.905047656
## 100            High         0.101736634          0.898263366
## 104            High         0.935621948          0.064378052
## 112            High         0.234756885          0.765243115
## 115            High         0.850921886          0.149078114
## 119            High         0.185157545          0.814842455
## 128            High         0.185157545          0.814842455
## 130            High         0.012185144          0.987814856
## 139            High         0.012185144          0.987814856
## 143            High         0.044803695          0.955196305
## 145            High         0.152326220          0.847673780
## 146            High         0.185157545          0.814842455
## 149            High         0.184218497          0.815781503
## 150            High         0.167743600          0.832256400
## 152            High         0.126560560          0.873439440
## 157            High         0.459326567          0.540673433
## 161            High         0.280613265          0.719386735
## 162            High         0.015166466          0.984833534
## 166            High         0.474765084          0.525234916
## 167            High         0.191503675          0.808496325
## 173            High         0.153251009          0.846748991
## 176            High         0.185157545          0.814842455
## 182            High         0.058745459          0.941254541
## 187            High         0.030012739          0.969987261
## 190            High         0.023048545          0.976951455
## 194            High         0.054829148          0.945170852
## 195            High         0.225743072          0.774256928
## 201            High         0.118943078          0.881056922
## 207            High         0.159359817          0.840640183
## 208            High         0.433804137          0.566195863
## 215            High         0.069295112          0.930704888
## 222            High         0.287429665          0.712570335
## 224            High         0.177837300          0.822162700
## 231            High         0.663173031          0.336826969
## 236            High         0.128407370          0.871592630
## 237            High         0.087117926          0.912882074
## 240            High         0.211568207          0.788431793
## 243            High         0.118943078          0.881056922
## 248            High         0.225743072          0.774256928
## 251            High         0.802510963          0.197489037
## 256            High         0.425809264          0.574190736
## 258            High         0.208604234          0.791395766
## 262            High         0.433804137          0.566195863
## 266            High         0.458916481          0.541083519
## 272            High         0.515097736          0.484902264
## 280            High         0.301261733          0.698738267
## 283            High         0.390880675          0.609119325
## 286            High         0.445587091          0.554412909
## 287            High         0.216476703          0.783523297
## 289            High         0.130497736          0.869502264
## 290            High         0.324389101          0.675610899
## 298            High         0.299117365          0.700882635
## 305            High         0.322034003          0.677965997
## 306            High         0.150503800          0.849496200
## 312            High         0.102017831          0.897982169
## 320            High         0.292893182          0.707106818
## 325            High         0.178390162          0.821609838
## 332            High         0.065417688          0.934582312
## 333            High         0.394656156          0.605343844
## 335            High         0.299117365          0.700882635
## 339            High         0.677983251          0.322016749
## 346            High         0.381040169          0.618959831
## 347            High         0.084143549          0.915856451
## 350            High         0.332513268          0.667486732
## 353            High         0.375699877          0.624300123
## 358            High         0.356002112          0.643997888
## 365            High         0.253020718          0.746979282
## 367            High         0.189824624          0.810175376
## 370            High         0.019857926          0.980142074
## 379            High         0.077838519          0.922161481
## 386            High         0.315641390          0.684358610
## 394            High         0.536219674          0.463780326
## 396            High         0.135699295          0.864300705
## 400            High         0.087117926          0.912882074
## 404            High         0.069987179          0.930012821
## 405            High         0.543420114          0.456579886
## 413            High         0.154491627          0.845508373
## 415            High         0.396880018          0.603119982
## 417            High         0.218484683          0.781515317
## 418            High         0.444542179          0.555457821
## 423            High         0.286471132          0.713528868
## 434            High         0.321591910          0.678408090
## 437            High         0.181118453          0.818881547
## 440            High         0.308562671          0.691437329
## 449            High         0.375209738          0.624790262
## 450            High         0.294263105          0.705736895
## 457            High         0.375209738          0.624790262
## 467            High         0.239925904          0.760074096
## 469            High         0.219694251          0.780305749
## 474            High         0.923897112          0.076102888
## 475            High         0.883349944          0.116650056
## 485            High         0.085422969          0.914577031
## 504             Low         0.291557950          0.708442050
## 511             Low         0.648063716          0.351936284
## 512             Low         0.433262622          0.566737378
## 517             Low         0.081862252          0.918137748
## 519             Low         0.768466561          0.231533439
## 520             Low         0.120730626          0.879269374
## 522             Low         0.925451743          0.074548257
## 527             Low         0.645300585          0.354699415
## 528             Low         0.271085584          0.728914416
## 529             Low         0.310908209          0.689091791
## 537             Low         0.160364901          0.839635099
## 540             Low         0.906505047          0.093494953
## 541             Low         0.445392323          0.554607677
## 547             Low         0.908345685          0.091654315
## 550             Low         0.648063716          0.351936284
## 555             Low         0.553749570          0.446250430
## 564             Low         0.076566109          0.923433891
## 570             Low         0.381921892          0.618078108
## 573             Low         0.263325647          0.736674353
## 575             Low         0.445392323          0.554607677
## 578             Low         0.200665198          0.799334802
## 581             Low         0.263325647          0.736674353
## 585             Low         0.328862698          0.671137302
## 590             Low         0.758709382          0.241290618
## 601             Low         0.954499169          0.045500831
## 602             Low         0.642143769          0.357856231
## 607             Low         0.609779376          0.390220624
## 610             Low         0.596902022          0.403097978
## 618             Low         0.770717139          0.229282861
## 624             Low         0.328862698          0.671137302
## 626             Low         0.264394512          0.735605488
## 627             Low         0.325327914          0.674672086
## 634             Low         0.859870031          0.140129969
## 640             Low         0.993333081          0.006666919
## 642             Low         0.183956965          0.816043035
## 643             Low         0.513477586          0.486522414
## 644             Low         0.822581654          0.177418346
## 645             Low         0.716527602          0.283472398
## 646             Low         0.656710655          0.343289345
## 647             Low         0.810727112          0.189272888
## 652             Low         0.207004348          0.792995652
## 658             Low         0.633864017          0.366135983
## 659             Low         0.756657553          0.243342447
## 660             Low         0.915400833          0.084599167
## 664             Low         0.317520700          0.682479300
## 666             Low         0.417931911          0.582068089
## 667             Low         0.877428595          0.122571405
## 675             Low         0.513477586          0.486522414
## 680             Low         0.994852845          0.005147155
## 681             Low         0.919786159          0.080213841
## 687             Low         0.785159445          0.214840555
## 694             Low         0.776291590          0.223708410
## 697             Low         0.504279816          0.495720184
## 701             Low         0.306165429          0.693834571
## 705             Low         0.986170693          0.013829307
## 707             Low         0.683050824          0.316949176
## 710             Low         0.639259875          0.360740125
## 716             Low         0.912237098          0.087762902
## 719             Low         0.933456382          0.066543618
## 720             Low         0.986060028          0.013939972
## 725             Low         0.989191020          0.010808980
## 727             Low         0.306165429          0.693834571
## 730             Low         0.468166324          0.531833676
## 738             Low         0.784293052          0.215706948
## 745             Low         0.580737162          0.419262838
## 748             Low         0.706782117          0.293217883
## 751             Low         0.980575777          0.019424223
## 756             Low         0.717150407          0.282849593
## 766             Low         0.850966796          0.149033204
## 769             Low         0.535328740          0.464671260
## 783             Low         0.811655480          0.188344520
## 785             Low         0.841947001          0.158052999
## 790             Low         0.923371376          0.076628624
## 793             Low         0.811655480          0.188344520
## 795             Low         0.995359452          0.004640548
## 796             Low         0.988726123          0.011273877
## 797             Low         0.613968875          0.386031125
## 801             Low         0.717464025          0.282535975
## 811             Low         0.486145842          0.513854158
## 812             Low         0.961781245          0.038218755
## 815             Low         0.940897618          0.059102382
## 816             Low         0.735876018          0.264123982
## 817             Low         0.926306993          0.073693007
## 824             Low         0.844929335          0.155070665
## 825             Low         0.844929335          0.155070665
## 826             Low         0.844929335          0.155070665
## 830             Low         0.855747468          0.144252532
## 837             Low         0.910197711          0.089802289
## 838             Low         0.735876018          0.264123982
## 844             Low         0.885768710          0.114231290
## 845             Low         0.976017309          0.023982691
## 847             Low         0.924759731          0.075240269
## 850             Low         0.874286786          0.125713214
## 852             Low         0.899365871          0.100634129
## 853             Low         0.899365871          0.100634129
## 861             Low         0.920297779          0.079702221
## 868             Low         0.980571498          0.019428502
## 874             Low         0.937576780          0.062423220
## 879            High         0.152866850          0.847133150
## 895            High         0.039547020          0.960452980
## 899            High         0.006274238          0.993725762
## 903            High         0.039547020          0.960452980
## 917            High         0.029052094          0.970947906
## 927            High         0.059988529          0.940011471
## 929            High         0.153251009          0.846748991
## 931            High         0.059988529          0.940011471
## 933            High         0.383904739          0.616095261
## 944            High         0.056057332          0.943942668
## 947            High         0.086098117          0.913901883
## 949            High         0.191869316          0.808130684
## 953            High         0.021217814          0.978782186
## 958            High         0.451442468          0.548557532
## 961            High         0.024500716          0.975499284
## 963            High         0.158030320          0.841969680
## 964            High         0.118943078          0.881056922
## 973            High         0.054972558          0.945027442
## 976            High         0.076566109          0.923433891
## 977            High         0.225743072          0.774256928
## 980            High         0.194325725          0.805674275
## 983            High         0.564399605          0.435600395
## 984            High         0.225743072          0.774256928
## 986            High         0.102017831          0.897982169
## 989            High         0.215977016          0.784022984
## 991            High         0.028136447          0.971863553
## 996            High         0.096066005          0.903933995
## 997            High         0.409217772          0.590782228
## 999            High         0.050174667          0.949825333
## 1000           High         0.049962301          0.950037699
## 1003           High         0.087117926          0.912882074
## 1008           High         0.110301043          0.889698957
## 1009           High         0.390880675          0.609119325
## 1014           High         0.054342796          0.945657204
## 1015           High         0.507479791          0.492520209
## 1040           High         0.173822737          0.826177263
## 1042           High         0.348406639          0.651593361
## 1043           High         0.664831519          0.335168481
## 1050           High         0.104964385          0.895035615
## 1052           High         0.217627219          0.782372781
## 1056           High         0.430563503          0.569436497
## 1070           High         0.576047259          0.423952741
## 1073           High         0.426680447          0.573319553
## 1074           High         0.199973894          0.800026106
## 1079           High         0.365577914          0.634422086
## 1080           High         0.495056564          0.504943436
## 1085           High         0.114818593          0.885181407
## 1087           High         0.677563755          0.322436245
## 1096           High         0.946509683          0.053490317
## 1099           High         0.476707448          0.523292552
## 1100           High         0.613540731          0.386459269
## 1102           High         0.037272700          0.962727300
## 1107            Low         0.378660320          0.621339680
## 1109            Low         0.746345913          0.253654087
## 1114            Low         0.404291532          0.595708468
## 1118            Low         0.392930424          0.607069576
## 1123            Low         0.449318373          0.550681627
## 1132            Low         0.827326214          0.172673786
## 1134            Low         0.625608459          0.374391541
## 1137            Low         0.325327914          0.674672086
## 1154            Low         0.325327914          0.674672086
## 1155            Low         0.600080216          0.399919784
## 1157            Low         0.848337304          0.151662696
## 1162            Low         0.513477586          0.486522414
## 1164            Low         0.183956965          0.816043035
## 1171            Low         0.980575777          0.019424223
## 1172            Low         0.459326567          0.540673433
## 1175            Low         0.709821865          0.290178135
## 1177            Low         0.578623299          0.421376701
## 1179            Low         0.955896251          0.044103749
## 1183            Low         0.250350289          0.749649711
## 1185            Low         0.971167041          0.028832959
## 1189            Low         0.957922810          0.042077190
## 1211            Low         0.714877986          0.285122014
## 1218            Low         0.994292098          0.005707902
## 1224            Low         0.372416312          0.627583688
## 1225            Low         0.306165429          0.693834571
## 1227            Low         0.950050219          0.049949781
## 1232            Low         0.993656325          0.006343675
## 1235            Low         0.777300182          0.222699818
## 1238            Low         0.805234045          0.194765955
## 1240            Low         0.900348204          0.099651796
## 1241            Low         0.714877986          0.285122014
## 1248            Low         0.855747468          0.144252532
## 1258            Low         0.735876018          0.264123982
## 1261            Low         0.881558385          0.118441615
## 1263            Low         0.844929335          0.155070665
## 1269            Low         0.961633901          0.038366099
## 1270            Low         0.995359452          0.004640548
## 1271            Low         0.874286786          0.125713214
## 1272            Low         0.899365871          0.100634129
## 1280            Low         0.937576780          0.062423220
## 1286            Low         0.983472472          0.016527528
## 1287            Low         0.985133774          0.014866226
## 1289            Low         0.962532737          0.037467263
## 1290            Low         0.899365871          0.100634129
## 1291           High         0.183663572          0.816336428
## 1294           High         0.737553964          0.262446036
## 1305            Low         0.961175729          0.038824271
## 1308           High         0.907626255          0.092373745

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_WT_Test_ROC <- roc(response = LR_WT_Test$LR_WT_Observed,
             predictor = LR_WT_Test$LR_WT_Predicted.High,
             levels = rev(levels(LR_WT_Test$LR_WT_Observed)))

(LR_WT_Test_ROCCurveAUC <- auc(LR_WT_Test_ROC)[1])

## [1] 0.8891629

1.5.10 Logistic Regression With Outlier Spatial Sign Treatment (LR_SST)

Logistic Regression models the relationship between the probability of an event (among two outcome levels) by having the log-odds of the event be a linear combination of a set of predictors weighted by their respective parameter estimates. The parameters are estimated via maximum likelihood estimation by testing different values through multiple iterations to optimize for the best fit of log odds. All of these iterations produce the log likelihood function, and logistic regression seeks to maximize this function to find the best parameter estimates. Given the optimal parameters, the conditional probabilities for each observation can be calculated, logged, and summed together to yield a predicted probability.

Spatial Sign Treatment is a multivariate extension of the concept of sign. The method projects the centered and scaled variables onto a unit sphere and is related to global contrast normalization. Spatial signs, when used as multivariate estimators of covariance structures are more robust to outlying observations.

[A] The logistic regression model from the stats package was implemented through the caret package with Spatial Sign treatment applied to treat data outliers but no any treatment applied for data skewness.

[B] The model does not contain any hyperparameter.

[C] The cross-validated model performance of the final model is summarized as follows:
     [C.1] Final model configuration is fixed due to the absence of a hyperparameter
     [C.2] AUROC = 0.86608

[D] The model allows for ranking of predictors in terms of variable importance. The top-performing predictors in the model are as follows:
     [D.1] HydrophilicFactor variable (numeric)
     [D.2] NumNonHAtoms variable (numeric)
     [D.3] NumAtoms variable (numeric)

[E] The independent test model performance of the final model is summarized as follows:
     [E.1] AUROC = 0.88786

Code Chunk | Output

##################################
# Creating a local object
# for the train and test sets
##################################
PMA_PreModelling_Train <- Solubility_Train
PMA_PreModelling_Test <- Solubility_Test

PMA_PreModelling_Train_LR <- PMA_PreModelling_Train
PMA_PreModelling_Test_LR <- PMA_PreModelling_Test

PMA_PreModelling_Train_LR.Numeric <- PMA_PreModelling_Train_LR[,sapply(PMA_PreModelling_Train_LR, is.numeric)]
PMA_PreModelling_Test_LR.Numeric <- PMA_PreModelling_Test_LR[,sapply(PMA_PreModelling_Test_LR, is.numeric)]

##################################
# Treating data skewness
# for the train set
##################################
# No actions applied

##################################
# Treating data outliers
# for the train set
##################################
Transform_SpatialSign <- preProcess(PMA_PreModelling_Train_LR, method = c("spatialSign"))
PMA_PreModelling_Train_LR_SST <- predict(Transform_SpatialSign, PMA_PreModelling_Train_LR.Numeric)
PMA_PreModelling_Train_LR_SST$Log_Solubility_Class <- PMA_PreModelling_Train_LR$Log_Solubility_Class

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Train_LR_SST[,sapply(PMA_PreModelling_Train_LR_SST, is.numeric)],
            y = PMA_PreModelling_Train_LR_SST$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_SST Train Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Train_LR_SST[,sapply(PMA_PreModelling_Train_LR_SST, is.numeric)], 
            y = PMA_PreModelling_Train_LR_SST$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_SST Train Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Train_LR_SST$Log_Solubility_Class)))))

##################################
# Creating consistent fold assignments
# for the 10-Fold Cross Validation process
##################################
set.seed(12345678)
KFold_Indices <- createFolds(PMA_PreModelling_Train_LR_SST$Log_Solubility_Class,
                             k = 10,
                             returnTrain=TRUE)
KFold_Control <- trainControl(method="cv",
                              index=KFold_Indices,
                              summaryFunction = twoClassSummary,
                              classProbs = TRUE)

##################################
# Setting the conditions
# for hyperparameter tuning
##################################
# No hyperparameter tuning process conducted
# hyperparameter=intercept fixed to TRUE

##################################
# Running the logistic regression model
# by setting the caret method to 'glm'
##################################
set.seed(12345678)
LR_SST_Tune <- train(x = PMA_PreModelling_Train_LR_SST[,!names(PMA_PreModelling_Train_LR_SST) %in% c("Log_Solubility_Class")],
                 y = PMA_PreModelling_Train_LR_SST$Log_Solubility_Class,
                 method = "glm",
                 metric = "ROC",
                 trControl = KFold_Control)

##################################
# Reporting the cross-validation results
# for the train set
##################################
LR_SST_Tune

## Generalized Linear Model 
## 
## 951 samples
##   4 predictor
##   2 classes: 'Low', 'High' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 857, 855, 857, 855, 856, 856, ... 
## Resampling results:
## 
##   ROC        Sens       Spec     
##   0.8660886  0.7375415  0.8493469

LR_SST_Tune$finalModel

## 
## Call:  NULL
## 
## Coefficients:
##       (Intercept)  HydrophilicFactor           NumAtoms       NumNonHAtoms  
##           0.17984            1.38112            0.05357           -1.58797  
##         NumCarbon  
##          -1.66022  
## 
## Degrees of Freedom: 950 Total (i.e. Null);  946 Residual
## Null Deviance:       1308 
## Residual Deviance: 868.7     AIC: 878.7

LR_SST_Tune$results

##   parameter       ROC      Sens      Spec      ROCSD     SensSD     SpecSD
## 1      none 0.8660886 0.7375415 0.8493469 0.04170766 0.06066318 0.05810749

(LR_SST_Train_ROCCurveAUC <- LR_SST_Tune$results$ROC)

## [1] 0.8660886

##################################
# Identifying and plotting the
# best model predictors
##################################
LR_SST_VarImp <- varImp(LR_SST_Tune, scale = TRUE)
plot(LR_SST_VarImp,
     top=4,
     scales=list(y=list(cex = .95)),
     main="Ranked Variable Importance : Logistic Regression",
     xlab="Scaled Variable Importance Metrics",
     ylab="Predictors",
     cex=2,
     origin=0,
     alpha=0.45)

##################################
# Treating data skewness
# for the test set
##################################
# No actions applied

##################################
# Treating data outliers
# for the test set
##################################
PMA_PreModelling_Test_LR_SST <- predict(Transform_SpatialSign, PMA_PreModelling_Test_LR.Numeric)
PMA_PreModelling_Test_LR_SST$Log_Solubility_Class <- PMA_PreModelling_Test_LR$Log_Solubility_Class

##################################
# Exploring the train set distribution
# of the numeric predictors
# with respect to the outcome
##################################
featurePlot(x = PMA_PreModelling_Test_LR_SST[,sapply(PMA_PreModelling_Test_LR_SST, is.numeric)],
            y = PMA_PreModelling_Test_LR_SST$Log_Solubility_Class,
            plot = "box",
            scales = list(x = list(relation="free", rot = 90),
                          y = list(relation="free")),
            adjust = 1.5,
            pch = "|", 
            main = "LR_SST Test Set : Numeric Predictor Distribution by Response Level")

featurePlot(x = PMA_PreModelling_Test_LR_SST[,sapply(PMA_PreModelling_Test_LR_SST, is.numeric)], 
            y = PMA_PreModelling_Test_LR_SST$Log_Solubility_Class,
            plot = "density",
            scales = list(x = list(relation="free", rot = 90), 
                          y = list(relation="free")),
            adjust = 1.5, 
            pch = "|", 
            main = "LR_SST Test Set : Numeric Predictor Distribution by Response Level",
            auto.key = list(columns = (length(levels(PMA_PreModelling_Test_LR_SST$Log_Solubility_Class)))))

##################################
# Independently evaluating the model
# on the test set
##################################
LR_SST_Test <- data.frame(LR_SST_Observed = PMA_PreModelling_Test_LR_SST$Log_Solubility_Class,
                      LR_SST_Predicted = predict(LR_SST_Tune,
                      PMA_PreModelling_Test_LR_SST[,!names(PMA_PreModelling_Test_LR_SST) %in% c("Log_Solubility_Class")],
                      type = "prob"))

LR_SST_Test

##      LR_SST_Observed LR_SST_Predicted.Low LR_SST_Predicted.High
## 20              High           0.09601045            0.90398955
## 21              High           0.07124817            0.92875183
## 23              High           0.08708570            0.91291430
## 25              High           0.06240783            0.93759217
## 28              High           0.08655612            0.91344388
## 31              High           0.13801964            0.86198036
## 32              High           0.16206633            0.83793367
## 33              High           0.16608858            0.83391142
## 34              High           0.16608858            0.83391142
## 37              High           0.26722180            0.73277820
## 38              High           0.26722180            0.73277820
## 42              High           0.88124946            0.11875054
## 49              High           0.18230992            0.81769008
## 54              High           0.09206998            0.90793002
## 55              High           0.08258681            0.91741319
## 58              High           0.24825896            0.75174104
## 60              High           0.13669151            0.86330849
## 61              High           0.16608858            0.83391142
## 65              High           0.10668377            0.89331623
## 69              High           0.30827773            0.69172227
## 73              High           0.11783250            0.88216750
## 86              High           0.15534315            0.84465685
## 90              High           0.09330688            0.90669312
## 91              High           0.09693780            0.90306220
## 93              High           0.09330688            0.90669312
## 96              High           0.09693780            0.90306220
## 98              High           0.10148441            0.89851559
## 100             High           0.12149523            0.87850477
## 104             High           0.85645369            0.14354631
## 112             High           0.21358439            0.78641561
## 115             High           0.88435693            0.11564307
## 119             High           0.17266138            0.82733862
## 128             High           0.17266138            0.82733862
## 130             High           0.08503124            0.91496876
## 139             High           0.08503124            0.91496876
## 143             High           0.05944353            0.94055647
## 145             High           0.18313278            0.81686722
## 146             High           0.17266138            0.82733862
## 149             High           0.13740764            0.86259236
## 150             High           0.18130010            0.81869990
## 152             High           0.16608858            0.83391142
## 157             High           0.26658480            0.73341520
## 161             High           0.17711690            0.82288310
## 162             High           0.07695031            0.92304969
## 166             High           0.41489928            0.58510072
## 167             High           0.19476630            0.80523370
## 173             High           0.17346217            0.82653783
## 176             High           0.17266138            0.82733862
## 182             High           0.14009484            0.85990516
## 187             High           0.09412624            0.90587376
## 190             High           0.08209356            0.91790644
## 194             High           0.13023242            0.86976758
## 195             High           0.16037985            0.83962015
## 201             High           0.09832548            0.90167452
## 207             High           0.12012692            0.87987308
## 208             High           0.34709451            0.65290549
## 215             High           0.13801964            0.86198036
## 222             High           0.20060734            0.79939266
## 224             High           0.09514653            0.90485347
## 231             High           0.66717290            0.33282710
## 236             High           0.12558408            0.87441592
## 237             High           0.15067576            0.84932424
## 240             High           0.14905340            0.85094660
## 243             High           0.09832548            0.90167452
## 248             High           0.16037985            0.83962015
## 251             High           0.83891581            0.16108419
## 256             High           0.48090891            0.51909109
## 258             High           0.14809516            0.85190484
## 262             High           0.34709451            0.65290549
## 266             High           0.27148151            0.72851849
## 272             High           0.42076835            0.57923165
## 280             High           0.37148413            0.62851587
## 283             High           0.33866037            0.66133963
## 286             High           0.25364415            0.74635585
## 287             High           0.15032543            0.84967457
## 289             High           0.14473001            0.85526999
## 290             High           0.36443155            0.63556845
## 298             High           0.24314628            0.75685372
## 305             High           0.17653728            0.82346272
## 306             High           0.24087855            0.75912145
## 312             High           0.09619891            0.90380109
## 320             High           0.25289581            0.74710419
## 325             High           0.13114586            0.86885414
## 332             High           0.05741702            0.94258298
## 333             High           0.33447896            0.66552104
## 335             High           0.24314628            0.75685372
## 339             High           0.76354018            0.23645982
## 346             High           0.37652804            0.62347196
## 347             High           0.13821283            0.86178717
## 350             High           0.32057761            0.67942239
## 353             High           0.26481814            0.73518186
## 358             High           0.36381440            0.63618560
## 365             High           0.15607953            0.84392047
## 367             High           0.12195706            0.87804294
## 370             High           0.09000562            0.90999438
## 379             High           0.14337494            0.85662506
## 386             High           0.27698113            0.72301887
## 394             High           0.74716571            0.25283429
## 396             High           0.08716053            0.91283947
## 400             High           0.15067576            0.84932424
## 404             High           0.13313956            0.86686044
## 405             High           0.50020364            0.49979636
## 413             High           0.10227313            0.89772687
## 415             High           0.24406283            0.75593717
## 417             High           0.22429372            0.77570628
## 418             High           0.66553672            0.33446328
## 423             High           0.23928941            0.76071059
## 434             High           0.24070864            0.75929136
## 437             High           0.23599992            0.76400008
## 440             High           0.43499212            0.56500788
## 449             High           0.54702006            0.45297994
## 450             High           0.15824323            0.84175677
## 457             High           0.54702006            0.45297994
## 467             High           0.32837858            0.67162142
## 469             High           0.18858281            0.81141719
## 474             High           0.87686350            0.12313650
## 475             High           0.87672351            0.12327649
## 485             High           0.28334254            0.71665746
## 504              Low           0.33447964            0.66552036
## 511              Low           0.60427805            0.39572195
## 512              Low           0.27333938            0.72666062
## 517              Low           0.13237259            0.86762741
## 519              Low           0.82383869            0.17616131
## 520              Low           0.16215286            0.83784714
## 522              Low           0.89837761            0.10162239
## 527              Low           0.79771009            0.20228991
## 528              Low           0.25999106            0.74000894
## 529              Low           0.49083183            0.50916817
## 537              Low           0.17724045            0.82275955
## 540              Low           0.89009261            0.10990739
## 541              Low           0.42590525            0.57409475
## 547              Low           0.90735268            0.09264732
## 550              Low           0.54923224            0.45076776
## 555              Low           0.76716356            0.23283644
## 564              Low           0.14548176            0.85451824
## 570              Low           0.61515621            0.38484379
## 573              Low           0.16732141            0.83267859
## 575              Low           0.42590525            0.57409475
## 578              Low           0.21645684            0.78354316
## 581              Low           0.16732141            0.83267859
## 585              Low           0.19579720            0.80420280
## 590              Low           0.89369063            0.10630937
## 601              Low           0.77618942            0.22381058
## 602              Low           0.83477322            0.16522678
## 607              Low           0.85385593            0.14614407
## 610              Low           0.86401175            0.13598825
## 618              Low           0.76648549            0.23351451
## 624              Low           0.19579720            0.80420280
## 626              Low           0.39079903            0.60920097
## 627              Low           0.24191991            0.75808009
## 634              Low           0.60202668            0.39797332
## 640              Low           0.85724614            0.14275386
## 642              Low           0.20979096            0.79020904
## 643              Low           0.56251635            0.43748365
## 644              Low           0.90646801            0.09353199
## 645              Low           0.82298391            0.17701609
## 646              Low           0.69510923            0.30489077
## 647              Low           0.74412785            0.25587215
## 652              Low           0.20093976            0.79906024
## 658              Low           0.78748187            0.21251813
## 659              Low           0.78880883            0.21119117
## 660              Low           0.86232533            0.13767467
## 664              Low           0.44522589            0.55477411
## 666              Low           0.21629092            0.78370908
## 667              Low           0.87244386            0.12755614
## 675              Low           0.56251635            0.43748365
## 680              Low           0.85164563            0.14835437
## 681              Low           0.86714281            0.13285719
## 687              Low           0.86569004            0.13430996
## 694              Low           0.91766599            0.08233401
## 697              Low           0.52480798            0.47519202
## 701              Low           0.31589620            0.68410380
## 705              Low           0.81924861            0.18075139
## 707              Low           0.77057558            0.22942442
## 710              Low           0.59761749            0.40238251
## 716              Low           0.82909104            0.17090896
## 719              Low           0.86938439            0.13061561
## 720              Low           0.85732525            0.14267475
## 725              Low           0.85641884            0.14358116
## 727              Low           0.31589620            0.68410380
## 730              Low           0.26046315            0.73953685
## 738              Low           0.83701684            0.16298316
## 745              Low           0.70271329            0.29728671
## 748              Low           0.79057047            0.20942953
## 751              Low           0.84112961            0.15887039
## 756              Low           0.84777333            0.15222667
## 766              Low           0.83906038            0.16093962
## 769              Low           0.29897447            0.70102553
## 783              Low           0.87595878            0.12404122
## 785              Low           0.89290405            0.10709595
## 790              Low           0.87181803            0.12818197
## 793              Low           0.87595878            0.12404122
## 795              Low           0.85791849            0.14208151
## 796              Low           0.85311015            0.14688985
## 797              Low           0.75584701            0.24415299
## 801              Low           0.53488474            0.46511526
## 811              Low           0.28134812            0.71865188
## 812              Low           0.84937551            0.15062449
## 815              Low           0.84844617            0.15155383
## 816              Low           0.68518843            0.31481157
## 817              Low           0.91238717            0.08761283
## 824              Low           0.89683201            0.10316799
## 825              Low           0.89683201            0.10316799
## 826              Low           0.89683201            0.10316799
## 830              Low           0.89256275            0.10743725
## 837              Low           0.90590054            0.09409946
## 838              Low           0.68518843            0.31481157
## 844              Low           0.90510276            0.09489724
## 845              Low           0.90252118            0.09747882
## 847              Low           0.92129203            0.07870797
## 850              Low           0.90778429            0.09221571
## 852              Low           0.91163676            0.08836324
## 853              Low           0.91163676            0.08836324
## 861              Low           0.91093261            0.08906739
## 868              Low           0.91054857            0.08945143
## 874              Low           0.90756578            0.09243422
## 879             High           0.16682367            0.83317633
## 895             High           0.09101760            0.90898240
## 899             High           0.08837534            0.91162466
## 903             High           0.09101760            0.90898240
## 917             High           0.09619149            0.90380851
## 927             High           0.09206998            0.90793002
## 929             High           0.17346217            0.82653783
## 931             High           0.09206998            0.90793002
## 933             High           0.51744048            0.48255952
## 944             High           0.08037883            0.91962117
## 947             High           0.09330688            0.90669312
## 949             High           0.10386583            0.89613417
## 953             High           0.07173711            0.92826289
## 958             High           0.45201022            0.54798978
## 961             High           0.05848371            0.94151629
## 963             High           0.15241824            0.84758176
## 964             High           0.09832548            0.90167452
## 973             High           0.07223382            0.92776618
## 976             High           0.14548176            0.85451824
## 977             High           0.16037985            0.83962015
## 980             High           0.33393088            0.66606912
## 983             High           0.40332216            0.59667784
## 984             High           0.16037985            0.83962015
## 986             High           0.09619891            0.90380109
## 989             High           0.20478385            0.79521615
## 991             High           0.06136135            0.93863865
## 996             High           0.15817455            0.84182545
## 997             High           0.29133806            0.70866194
## 999             High           0.08708570            0.91291430
## 1000            High           0.08096203            0.91903797
## 1003            High           0.15067576            0.84932424
## 1008            High           0.08653032            0.91346968
## 1009            High           0.33866037            0.66133963
## 1014            High           0.06211801            0.93788199
## 1015            High           0.68254254            0.31745746
## 1040            High           0.10197668            0.89802332
## 1042            High           0.36045604            0.63954396
## 1043            High           0.62172343            0.37827657
## 1050            High           0.07340589            0.92659411
## 1052            High           0.15785081            0.84214919
## 1056            High           0.22781432            0.77218568
## 1070            High           0.82152666            0.17847334
## 1073            High           0.26265879            0.73734121
## 1074            High           0.19955598            0.80044402
## 1079            High           0.33568619            0.66431381
## 1080            High           0.60786098            0.39213902
## 1085            High           0.14911681            0.85088319
## 1087            High           0.89957835            0.10042165
## 1096            High           0.85587597            0.14412403
## 1099            High           0.68096864            0.31903136
## 1100            High           0.69132245            0.30867755
## 1102            High           0.09812815            0.90187185
## 1107             Low           0.33306906            0.66693094
## 1109             Low           0.82831781            0.17168219
## 1114             Low           0.34072119            0.65927881
## 1118             Low           0.60266074            0.39733926
## 1123             Low           0.51544312            0.48455688
## 1132             Low           0.89086611            0.10913389
## 1134             Low           0.86196038            0.13803962
## 1137             Low           0.24191991            0.75808009
## 1154             Low           0.24191991            0.75808009
## 1155             Low           0.82419178            0.17580822
## 1157             Low           0.79195710            0.20804290
## 1162             Low           0.56251635            0.43748365
## 1164             Low           0.20979096            0.79020904
## 1171             Low           0.84112961            0.15887039
## 1172             Low           0.26658480            0.73341520
## 1175             Low           0.81770909            0.18229091
## 1177             Low           0.73170491            0.26829509
## 1179             Low           0.86003399            0.13996601
## 1183             Low           0.25318113            0.74681887
## 1185             Low           0.88391319            0.11608681
## 1189             Low           0.77072878            0.22927122
## 1211             Low           0.80815455            0.19184545
## 1218             Low           0.87884779            0.12115221
## 1224             Low           0.24958622            0.75041378
## 1225             Low           0.31589620            0.68410380
## 1227             Low           0.87910711            0.12089289
## 1232             Low           0.85531787            0.14468213
## 1235             Low           0.90784054            0.09215946
## 1238             Low           0.80846734            0.19153266
## 1240             Low           0.88005225            0.11994775
## 1241             Low           0.80815455            0.19184545
## 1248             Low           0.89256275            0.10743725
## 1258             Low           0.68518843            0.31481157
## 1261             Low           0.86946402            0.13053598
## 1263             Low           0.89683201            0.10316799
## 1269             Low           0.89945602            0.10054398
## 1270             Low           0.84133006            0.15866994
## 1271             Low           0.90778429            0.09221571
## 1272             Low           0.91163676            0.08836324
## 1280             Low           0.90756578            0.09243422
## 1286             Low           0.90492295            0.09507705
## 1287             Low           0.90812642            0.09187358
## 1289             Low           0.89739056            0.10260944
## 1290             Low           0.91163676            0.08836324
## 1291            High           0.09497549            0.90502451
## 1294            High           0.86886342            0.13113658
## 1305             Low           0.86138517            0.13861483
## 1308            High           0.75006281            0.24993719

##################################
# Reporting the independent evaluation results
# for the test set
##################################
LR_SST_Test_ROC <- roc(response = LR_SST_Test$LR_SST_Observed,
             predictor = LR_SST_Test$LR_SST_Predicted.High,
             levels = rev(levels(LR_SST_Test$LR_SST_Observed)))

(LR_SST_Test_ROCCurveAUC <- auc(LR_SST_Test_ROC)[1])

## [1] 0.8878694

1.6 Consolidated Findings

[A] The transformation for skewed data generally performed better for models which are sensitive to deviations against the normality assumption (i.e. logistic regression) as compared to treatment for extreme outliers.
     [A.1] LR: Logistic Regression (stats package)
            [A.1.1] LR_REF (Without Skewness and Outlier Treatment):
                     [A.1.1.1] Cross-Validation AUROC = 0.87475
                     [A.1.1.2] Test AUROC = 0.88447
            [A.1.2] LR_BCT (With Box-Cox Transformation):
                     [A.1.2.1] Cross-Validation AUROC = 0.88878
                     [A.1.2.2] Test AUROC = 0.89676
            [A.1.3] LR_YJT (With Yeo-Johnson Transformation):
                     [A.1.3.1] Cross-Validation AUROC = 0.88070
                     [A.1.3.2] Test AUROC = 0.89061
            [A.1.4] LR_ET (With Exponential Transformation):
                     [A.1.4.1] Cross-Validation AUROC = 0.88053
                     [A.1.4.2] Test AUROC = 0.89013
            [A.1.5] LR_IHST (With Inverse Hyperbolic Sine Transformation):
                     [A.1.5.1] Cross-Validation AUROC = 0.87107
                     [A.1.5.2] Test AUROC = 0.87998
            [A.1.6] LR_LOG10T (With Base-10 Logarithm Transformation):
                     [A.1.6.1] Cross-Validation AUROC = 0.89192
                     [A.1.6.2] Test AUROC = 0.89882
            [A.1.7] LR_LNT (With Natural Logarithm Transformation):
                     [A.1.7.1] Cross-Validation AUROC = 0.89192
                     [A.1.7.2] Test AUROC = 0.89882
            [A.1.8] LR_SRT (With Square Root Transformation):
                     [A.1.8.1] Cross-Validation AUROC = 0.88441
                     [A.1.8.2] Test AUROC = 0.89461
            [A.1.9] LR_WT (With Outlier Winsorization Treatment):
                     [A.1.9.1] Cross-Validation AUROC = 0.87428
                     [A.1.9.2] Test AUROC = 0.88916
            [A.1.10] LR_SST (With Outlier Spatial Sign Treatment):
                     [A.1.10.1] Cross-Validation AUROC = 0.86608
                     [A.1.10.2] Test AUROC = 0.88786

Code Chunk | Output

##################################
# Consolidating all evaluation results
# for the train and test sets
# using the AUROC metric
##################################
Model <- c('LR_REF','LR_BCT','LR_YJT','LR_ET','LR_IHST','LR_LOG10T','LR_LNT','LR_SRT','LR_WT','LR_SST',
           'LR_REF','LR_BCT','LR_YJT','LR_ET','LR_IHST','LR_LOG10T','LR_LNT','LR_SRT','LR_WT','LR_SST')

Set <- c(rep('Cross-Validation',10),rep('Test',10))

ROCCurveAUC <- c(LR_Train_ROCCurveAUC,LR_BCT_Train_ROCCurveAUC,
                 LR_YJT_Train_ROCCurveAUC,LR_ET_Train_ROCCurveAUC,
                 LR_IHST_Train_ROCCurveAUC,LR_LOG10T_Train_ROCCurveAUC,
                 LR_LNT_Train_ROCCurveAUC,LR_SRT_Train_ROCCurveAUC,
                 LR_WT_Train_ROCCurveAUC,LR_SST_Train_ROCCurveAUC,
                 LR_Test_ROCCurveAUC,LR_BCT_Test_ROCCurveAUC,
                 LR_YJT_Test_ROCCurveAUC,LR_ET_Test_ROCCurveAUC,
                 LR_IHST_Test_ROCCurveAUC,LR_LOG10T_Test_ROCCurveAUC,
                 LR_LNT_Test_ROCCurveAUC,LR_SRT_Test_ROCCurveAUC,
                 LR_WT_Test_ROCCurveAUC,LR_SST_Test_ROCCurveAUC)

ROCCurveAUC_Summary <- as.data.frame(cbind(Model,Set,ROCCurveAUC))

ROCCurveAUC_Summary$ROCCurveAUC <- as.numeric(as.character(ROCCurveAUC_Summary$ROCCurveAUC))
ROCCurveAUC_Summary$Set <- factor(ROCCurveAUC_Summary$Set,
                                        levels = c("Cross-Validation",
                                                   "Test"))
ROCCurveAUC_Summary$Model <- factor(ROCCurveAUC_Summary$Model,
                                        levels = c('LR_REF',
                                                   'LR_BCT',
                                                   'LR_YJT',
                                                   'LR_ET',
                                                   'LR_IHST',
                                                   'LR_LOG10T',
                                                   'LR_LNT',
                                                   'LR_SRT',
                                                   'LR_WT',
                                                   'LR_SST'))

print(ROCCurveAUC_Summary, row.names=FALSE)

##      Model              Set ROCCurveAUC
##     LR_REF Cross-Validation   0.8747542
##     LR_BCT Cross-Validation   0.8887838
##     LR_YJT Cross-Validation   0.8807066
##      LR_ET Cross-Validation   0.8805333
##    LR_IHST Cross-Validation   0.8710722
##  LR_LOG10T Cross-Validation   0.8919210
##     LR_LNT Cross-Validation   0.8919210
##     LR_SRT Cross-Validation   0.8844149
##      LR_WT Cross-Validation   0.8742874
##     LR_SST Cross-Validation   0.8660886
##     LR_REF             Test   0.8844739
##     LR_BCT             Test   0.8967622
##     LR_YJT             Test   0.8906181
##      LR_ET             Test   0.8901330
##    LR_IHST             Test   0.8799871
##  LR_LOG10T             Test   0.8988237
##     LR_LNT             Test   0.8988237
##     LR_SRT             Test   0.8946198
##      LR_WT             Test   0.8891629
##     LR_SST             Test   0.8878694

(ROCCurveAUC_Plot <- dotplot(Model ~ ROCCurveAUC,
                           data = ROCCurveAUC_Summary,
                           groups = Set,
                           main = "Classification Model Performance Comparison",
                           ylab = "Model",
                           xlab = "AUROC",
                           auto.key = list(adj=1, space="top", columns=2),
                           type=c("p", "h"),
                           origin = 0,
                           alpha = 0.45,
                           pch = 16,
                           cex = 2))

2. Summary

3. References

[Book] Applied Predictive Modeling by Max Kuhn and Kjell Johnson
[Book] An Introduction to Statistical Learning by Gareth James, Daniela Witten, Trevor Hastie and Rob Tibshirani
[Book] Multivariate Data Visualization with R by Deepayan Sarkar
[Book] Machine Learning by Samuel Jackson
[Book] Data Modeling Methods by Jacob Larget
[Book] Introduction to R and Statistics by University of Western Australia
[Book] Feature Engineering and Selection: A Practical Approach for Predictive Models by Max Kuhn and Kjell Johnson
[Book] Introduction to Research Methods by Eric van Holm
[R Package] AppliedPredictiveModeling by Max Kuhn
[R Package] caret by Max Kuhn
[R Package] rpart by Terry Therneau and Beth Atkinson
[R Package] lattice by Deepayan Sarkar
[R Package] dplyr by Hadley Wickham
[R Package] moments by Lukasz Komsta and Frederick
[R Package] skimr by Elin Waring
[R Package] RANN by Sunil Arya, David Mount, Samuel Kemp and Gregory Jefferis
[R Package] corrplot by Taiyun Wei
[R Package] tidyverse by Hadley Wickham
[R Package] lares by Bernardo Lares
[R Package] DMwR2 by Luis Torgo
[R Package] gridExtra by Baptiste Auguie and Anton Antonov
[R Package] rattle by Graham Williams
[R Package] rpart.plot by Stephen Milborrow
[R Package] RColorBrewer by Erich Neuwirth
[R Package] stats by R Core Team
[R Package] pls by Kristian Hovde Liland
[R Package] nnet by Brian Ripley
[R Package] elasticnet by Hui Zou
[R Package] earth by Stephen Milborrow
[R Package] party by Torsten Hothorn
[R Package] kernlab by Alexandros Karatzoglou
[R Package] randomForest by Andy Liaw
[R Package] pROC by Xavier Robin
[R Package] mda by Trevor Hastie
[R Package] klaR by Christian Roever, Nils Raabe, Karsten Luebke, Uwe Ligges, Gero Szepannek, Marc Zentgraf and David Meyer
[R Package] pamr by Trevor Hastie, Rob Tibshirani, Balasubramanian Narasimhan and Gil Chu
[Article] The caret Package by Max Kuhn
[Article] A Short Introduction to the caret Package by Max Kuhn
[Article] Caret Package – A Practical Guide to Machine Learning in R by Selva Prabhakaran
[Article] Tuning Machine Learning Models Using the Caret R Package by Jason Brownlee
[Article] Lattice Graphs by Alboukadel Kassambara
[Article] A Tour of Machine Learning Algorithms by Jason Brownlee
[Article] 4 Types of Classification Tasks in Machine Learning by Jason Brownlee
[Article] Transforming Skewed Data: How to Choose the Right Transformation for your Distribution by Anatomise Biostats Team
[Article] Handling Outlying Or Skewed Data With Robust Regression by Joanna Diong
[Article] What’s Wrong With Your Statistical Model? Skewed Data. by Rajat Sharma
[Article] Winsorize: Definition, Examples in Easy Steps by Statistics How To Team
[Article] Winsorization: The Good, The Bad, and The Ugly by Rick Wicklin
[Article] 5 Variable Transformations to Improve Your Regression Model by George Choueiry
[Article] What is a Yeo-Johnson Power Transformation? by Ready Signal Team
[Article] Box-Cox Transformation in Excel (Step-by-Step) by Statology Team
[Article] Box-Cox Transformation by SPC For Excel Team
[Article] Box-Cox Transformation by Encyclopedia of Math Team
[Article] The Box-Cox Transformation: What It Is and How to Use It by Leanscape Team
[Article] Inverse Hyperbolic Sine by Wolfram Mathworld Team
[Article] Spot-Check Classification Machine Learning Algorithms in Python with scikit-learn by Jason Brownlee
[Article] Feature Engineering and Selection: A Practical Approach for Predictive Models by Max Kuhn and Kjell Johnson
[Article] Generalized Linear Model by BCCVL Team
[Article] Transformations: An Introduction by Boston College
[Publication] The Origins of Logistic Regression by JS Cramer (Econometrics eJournal)
[Publication] An Analysis of Transformations by George Box and David Cox (Journal of the Royal Statistical Society)
[Publication] A New Family of Power Transformations to Improve Normality or Symmetry by In-Kwon Yeo and Richard Johnson (Biometrika)
[Publication] Spatial Sign Preprocessing: A Simple Way To Impart Moderate Robustness to Multivariate Estimators by Sven Serneels, Evert De Nolf and Pierre Van Espen (Journal of Chemical Information and Modeling)
[Publication] Exponential Data Transformations by B Manly (Journal of the Royal Statistical Society)
[Publication] Notes on the Use of Data Transformations by Jason Osborne (Practical Assessment, Research, and Evaluation)
[Publication] Winsorization for Identifying and Treating Outliers in Business Surveys by Ray Chambers, Philip Kokic, Paul Smith and Marie Cruddas (Proceedings in the Second International Conference on Establishment Survey)
[Publication] The Inverse Hyperbolic Sine Transformation and Retransformed Marginal Effects by Edward Norton (National Bureau of Economic Research)
[Publication] An Application of the Inverse Hyperbolic Sine Transformation — A Note by Mingliang Zhang, John Fortney, John Tilford and Kathryn Rost (Health Services and Outcomes Research Methodology)
[Course] Applied Data Mining and Statistical Learning by Penn State Eberly College of Science
[Course] Regression Methods by Penn State Eberly College of Science
[Course] Applied Regression Analysis by Penn State Eberly College of Science