Decision Tree (Information Gain)

LOADING DATA INTO R ENVIRONMENT

Reading the Data

# reading the data
library(data.table)
df <- fread("MCICreditCardDefault.csv")

Converting some data columns into factors

# convert 'Male' as a factor
df[, Male := as.factor(Male)]
# convert 'Education' as a factor
df[, Education := as.factor(Education)]
# convert 'MaritalStatus' as a factor
df[, MaritalStatus := as.factor(MaritalStatus)]
# convert 'Default' as a factor
df[, Default := as.factor(Default)]

## setting levels as "Yes" and "NO"
df$Default <- ifelse(df$Default == "1","Yes","No")

Splitting data into training and test set

library(caret)
# data partition
set.seed(2341)
trainIndex <- createDataPartition(df$Default, p = 0.80, list = FALSE)
# 80% training data
train.df <- df[trainIndex, ]
# 20% testing data
test.df <- df[-trainIndex, ]
dim(train.df)

## [1] 23681     9

dim(test.df)

## [1] 5920    9

TRAINING THE DECISION TREE MODEL

Setting Control parameters

library(caret)
# control parameters
trctrl <- trainControl(method = "cv", n = 10, classProbs = TRUE)

Running the Training Model

set.seed(3333)
# fitting decision tree classification model
DTModel <- train(Default ~ CreditLimit
                         + Male
                         + Education
                         + MaritalStatus
                         + Age
                         + BillOutstanding
                         + LastPayment, 
                         data = train.df, 
                         method = "rpart",
                         metric = "ROC",
                         parms  = list(split = "information"), 
                         trControl = trctrl)
# model summary
DTModel

## CART 
## 
## 23681 samples
##     7 predictor
##     2 classes: 'No', 'Yes' 
## 
## No pre-processing
## Resampling: Cross-Validated (10 fold) 
## Summary of sample sizes: 21313, 21314, 21313, 21313, 21312, 21314, ... 
## Resampling results across tuning parameters:
## 
##   cp           Accuracy   Kappa     
##   0.001703255  0.7776698  0.05668749
##   0.002081756  0.7777964  0.05365759
##   0.003406510  0.7765299  0.01096180
## 
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was cp = 0.002081756.

Visualizing Tree using package rpart.plot

# viasulaziation
library(rpart.plot)
prp(DTModel$finalModel, box.palette = "Reds", tweak = 1.2, varlen = 20)

Variable Importance in Decision Tree Model

# plotting variable importance
plot(varImp(DTModel))

TESTING THE DECISION TREE MODEL

Predicting Model on Test Data Set

# predicting the model on test data set
PredDTModel <- predict(DTModel, 
                    newdata = test.df,type = "prob")

Plotting the Predicted Probabilities

# plot of probabilities
plot(PredDTModel$Yes, 
     main = "Scatterplot of Probabilities of default (test data)", 
     xlab = "Customer ID", 
     ylab = "Predicted Probability of default")

Confusion Matrix at 50% Cut-Off Probability

# taking the cut-off probability 50%
Predicted <- ifelse(PredDTModel$Yes > 0.50, "Yes", "No")

# ordering the vectors
Actual <- test.df$Default
# making confusion matrix
cm <-confusionMatrix(table(Predicted,Actual))
cm

## Confusion Matrix and Statistics
## 
##          Actual
## Predicted   No  Yes
##       No  4533 1234
##       Yes   66   87
##                                           
##                Accuracy : 0.7804          
##                  95% CI : (0.7696, 0.7909)
##     No Information Rate : 0.7769          
##     P-Value [Acc > NIR] : 0.2617          
##                                           
##                   Kappa : 0.0752          
##                                           
##  Mcnemar's Test P-Value : <2e-16          
##                                           
##             Sensitivity : 0.98565         
##             Specificity : 0.06586         
##          Pos Pred Value : 0.78602         
##          Neg Pred Value : 0.56863         
##              Prevalence : 0.77686         
##          Detection Rate : 0.76571         
##    Detection Prevalence : 0.97416         
##       Balanced Accuracy : 0.52575         
##                                           
##        'Positive' Class : No              
## 

ROC Curve using Package ROCR

# loading the package
library(ROCR)
DTPrediction <- predict(DTModel, test.df,type = "prob")
Prediction <- prediction(DTPrediction[2],test.df$Default)
performance <- performance(Prediction, "tpr","fpr")
# plotting ROC curve
plot(performance,main = "ROC Curve",col = 2,lwd = 2)
abline(a = 0,b = 1,lwd = 2,lty = 3,col = "black")

AUC (Area Under the Curve)

library(ROCR)
# area under curve
DTPrediction <- prediction(PredDTModel[2],test.df$Default)
aucDT <- performance(DTPrediction, measure = "auc")
aucDT <- aucDT@y.values[[1]]
aucDT

## [1] 0.601756