Classification Based on Decision Tree

---

# reading data
library(data.table)
autoF.dt <- fread("AutoFinanaceData.csv")
# attach data columns
attach(autoF.dt)
# dimension of the dataset
dim(autoF.dt)

[1] 28906    21

# structure of the data table
str(autoF.dt)

# descriptive statistics of the dataframe
library(psych)
describe(autoF.dt)[, c(1:5)]

Reserve 80% for training and 20% of test

# loading the package
library(caTools)
# fixing the observations 
set.seed(123)
# splitting the data 
split = sample.split(autoF.dt$DefaulterFlag, SplitRatio = 0.75)
# creating the training set
trainingSet = subset(autoF.dt, split == TRUE)
# creating the test set
testSet = subset(autoF.dt, split == FALSE)

# dimensions of the full data
dim(autoF.dt)

[1] 28906    21

# dimensions of the training, testing data
dim(trainingSet)

[1] 21679    21

dim(testSet)

[1] 7227   21

library(caret)
# Control Parameters to be used in the Decision Tree
ctrl <- trainControl(method = "cv", 
                     number = 8,
                     classProbs = TRUE, 
                     summaryFunction = twoClassSummary)

# Technical Details
# method :  resampling method used, e.g. {"boot", "optimism_boot", "boot_all", "cv", "repeatedcv", "LOOCV", "LGOCV"}
# number    : number of folds 
# summaryFunction: compute performance metrics across resamples. 

set.seed(2345)
library(caret)
levels(trainingSet$DefaulterFlag) <- c("No", "Yes")
dTreeInfoGain <- train(DefaulterFlag ~ AGE
                            + NOOFDEPE
                            + MTHINCTH 
                            + SALDATFR 
                            + TENORYR
                            + DWNPMFR
                            + PROFBUS
                            + QUALHSC 
                            + QUAL_PG
                            + SEXCODE
                            + FULLPDC
                            + FRICODE
                            + WASHCODE
                            + Region, 
                         data = trainingSet, 
                         method = "rpart", 
                         parms = list(split = "information"), 
                         trControl = ctrl,
                         metric = "ROC")

CART 

21679 samples
   14 predictor
    2 classes: 'No', 'Yes' 

No pre-processing
Resampling: Cross-Validated (8 fold) 
Summary of sample sizes: 18970, 18969, 18969, 18969, 18969, 18969, ... 
Resampling results across tuning parameters:

  cp          ROC        Sens       Spec     
  0.01024164  0.6782253  0.2632497  0.9163298
  0.01152184  0.6779262  0.2550870  0.9187274
  0.01408225  0.6337774  0.1922215  0.9371999

ROC was used to select the optimal model using the largest value.
The final value used for the model was cp = 0.01024164.

dTreeInfoGain$bestTune

          cp
1 0.01024164

• cp is the “minimum benefit” that a split must add to the tree. It is used to control the size of the decision tree and to select the optimal tree size.

• If the cost of adding another variable to the decision tree from the current node is above the value of cp, then tree building stops.

# plot complexity parameter
plot(dTreeInfoGain)

Recall that the data has multiple explanatory variables.

colnames(autoF.dt)

 [1] "Agmt No"        "ContractStatus" "StartDate"      "AGE"           
 [5] "NOOFDEPE"       "MTHINCTH"       "SALDATFR"       "TENORYR"       
 [9] "DWNPMFR"        "PROFBUS"        "QUALHSC"        "QUAL_PG"       
[13] "SEXCODE"        "FULLPDC"        "FRICODE"        "WASHCODE"      
[17] "Region"         "Branch"         "DefaulterFlag"  "DefaulterType" 
[21] "DATASET"       

Which ones are the most important?

# important variables
plot(varImp(dTreeInfoGain, 
            main = "Important Variables",
            scale = TRUE))

Segment 1:

Consumers who do not submit PDC (Post-dated Checks)

Segment 2:

Consumers who (i) submit PDC, (ii) Loan Tenure >= 1.4 years

Segment 3:

Consumers who (i) submit PDC; (ii) Loan Tenure < 1.4 years; (iii) Don't live in Chennai; (iv) Down-Payment < 37% of Loan

Segment 4:

Consumers who (i) submit PDC; (ii) Loan Tenure < 1.4 years; (iii) Don't live in Chennai; (iv) Down-Payment >= 37% of Loan; (v) Age < 39 years

# gini index
library(caret)
levels(trainingSet$DefaulterFlag) <- c("No", "Yes")
dTreeGiniIndex <- train(DefaulterFlag ~ AGE
                            + NOOFDEPE
                            + MTHINCTH 
                            + SALDATFR 
                            + TENORYR
                            + DWNPMFR
                            + PROFBUS
                            + QUALHSC 
                            + QUAL_PG
                            + SEXCODE
                            + FULLPDC
                            + FRICODE
                            + WASHCODE
                            + Region, 
                         data = trainingSet, 
                         method = "rpart", 
                         parms = list(split = "gini"), 
                         trControl = ctrl,
                         metric = "ROC")

CART 

21679 samples
   14 predictor
    2 classes: 'No', 'Yes' 

No pre-processing
Resampling: Cross-Validated (8 fold) 
Summary of sample sizes: 18970, 18969, 18969, 18970, 18968, 18969, ... 
Resampling results across tuning parameters:

  cp          ROC        Sens       Spec     
  0.01024164  0.6784573  0.2643651  0.9162687
  0.01152184  0.6781444  0.2800501  0.9077798
  0.01408225  0.6557567  0.2642051  0.9101078

ROC was used to select the optimal model using the largest value.
The final value used for the model was cp = 0.01024164.

# plot complexity parameter
plot(dTreeGiniIndex)

# prediction on test data
predClassTestInfoGain <- predict(dTreeInfoGain, 
                         testSet, 
                         type = 'raw')
levels(testSet$DefaulterFlag) <- c("No", "Yes")
# confusion matrix
confusionMatrix(predClassTestInfoGain, testSet$DefaulterFlag, 
                positive = "Yes")

predProbTestInfoGain <- predict(dTreeInfoGain, testSet, type = "prob")
library(ROCR)
lgPredObjtree <- prediction(predProbTestInfoGain[2],testSet$DefaulterFlag)
lgPerfObjtree <- performance(lgPredObjtree, "tpr","fpr")
plot(lgPerfObjtree,main = "ROC Curve",col = 2,lwd = 2)
abline(a = 0,b = 1,lwd = 2,lty = 3,col = "black")

# auc for decision tree
aucTree <- performance(lgPredObjtree, measure = "auc")
aucTree <- aucTree@y.values[[1]]
aucTree

[1] 0.6829464

CART 

21679 samples
   14 predictor
    2 classes: 'No', 'Yes' 

No pre-processing
Resampling: Cross-Validated (8 fold) 
Summary of sample sizes: 18970, 18969, 18969, 18969, 18969, 18969, ... 
Resampling results across tuning parameters:

  cp          ROC        Sens       Spec     
  0.01024164  0.6782253  0.2632497  0.9163298
  0.01152184  0.6779262  0.2550870  0.9187274
  0.01408225  0.6337774  0.1922215  0.9371999

ROC was used to select the optimal model using the largest value.
The final value used for the model was cp = 0.01024164.

Segment 1:

Consumers who do not submit PDC (Post-dated Checks)

Segment 2:

Consumers who (i) submit PDC, (ii) Loan Tenure >= 1.4 years

Segment 3:

Consumers who (i) submit PDC; (ii) Loan Tenure < 1.4 years; (iii) Don't live in Chennai; (iv) Down-Payment < 37% of Loan

Segment 4:

Consumers who (i) submit PDC; (ii) Loan Tenure < 1.4 years; (iii) Don't live in Chennai; (iv) Down-Payment >= 37% of Loan; (v) Age < 39 years

# auc for decision tree
aucTree <- performance(lgPredObjtree, measure = "auc")
aucTree <- aucTree@y.values[[1]]
aucTree

[1] 0.6829464