Bagging

LOADING DATA INTO R ENVIRONMENT

Reading the Data

# reading the data
df <- read.csv("DefaultData.csv")

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] 8001    4

dim(test.df)

## [1] 1999    4

TRAINING THE BAGGING MODEL

Setting Control parameters

library(caret)
# control parameters
trctrl <- trainControl(method = "none",classProbs = TRUE)

Running the Training Model

set.seed(123)
# fitting random forest classification model
BaggingModel <- train(default ~ ., 
                         data = train.df, 
                         method = "treebag",
                         nbagg = 50,
                         trControl = trctrl)
# model summary
BaggingModel

## Bagged CART 
## 
## 8001 samples
##    3 predictor
##    2 classes: 'No', 'Yes' 
## 
## No pre-processing
## Resampling: None

Variable Importance in Bagging Model

# plotting variable importance
plot(varImp(BaggingModel))

TESTING THE BAGGING MODEL

Predicting Model on Test Data Set

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

Plotting the Predicted Probabilities

# plot of probabilities
plot(PredBagModel$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%
PredBagModel <- ifelse(PredBagModel$Yes > 0.50, "Yes", "No")

# saving predicted vector as factor 
Pred <- as.factor(PredBagModel)

# ordering the vectors
Predicted <- ordered(Pred, levels = c("Yes", "No"))
Actual <- ordered(test.df$default,levels = c("Yes", "No"))

# making confusion matrix
cm <-confusionMatrix(data = Predicted,reference = Actual, positive = "Yes")
cm

## Confusion Matrix and Statistics
## 
##           Reference
## Prediction  Yes   No
##        Yes   29   16
##        No    37 1917
##                                           
##                Accuracy : 0.9735          
##                  95% CI : (0.9655, 0.9801)
##     No Information Rate : 0.967           
##     P-Value [Acc > NIR] : 0.05520         
##                                           
##                   Kappa : 0.5094          
##                                           
##  Mcnemar's Test P-Value : 0.00601         
##                                           
##             Sensitivity : 0.43939         
##             Specificity : 0.99172         
##          Pos Pred Value : 0.64444         
##          Neg Pred Value : 0.98106         
##              Prevalence : 0.03302         
##          Detection Rate : 0.01451         
##    Detection Prevalence : 0.02251         
##       Balanced Accuracy : 0.71556         
##                                           
##        'Positive' Class : Yes             
## 

ROC Curve using Package ROCR

# loading the package
library(ROCR)

## Warning: package 'ROCR' was built under R version 4.0.4

BagPrediction <- predict(BaggingModel, test.df,type = "prob")
BagPrediction <- prediction(BagPrediction[2],test.df$default)
Bagperformance <- performance(BagPrediction, "tpr","fpr")
# plotting ROC curve
plot(Bagperformance,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
aucBag <- performance(BagPrediction, measure = "auc")
aucBag <- aucBag@y.values[[1]]
aucBag

## [1] 0.9061123