Taiwan KNN
Sabarish M
July 6,2019
#PART 1: Read the data
library(data.table)
# reading data as data.table
CCdefault.dt <- fread("MCICreditCardDefault.csv")
# attaching the data
attach(CCdefault.dt)
# dimension of the data table
dim(CCdefault.dt)## [1] 29601 9# column names
colnames(CCdefault.dt)## [1] "Id" "CreditLimit" "Male" "Education"
## [5] "MaritalStatus" "Age" "BillOutstanding" "LastPayment"
## [9] "Default"#PART 2: Verifying data type structure
# structure of the dataframe
str(CCdefault.dt)## Classes 'data.table' and 'data.frame': 29601 obs. of 9 variables:
## $ Id : int 1 2 3 4 5 6 7 8 9 10 ...
## $ CreditLimit : int 20000 120000 90000 50000 50000 50000 500000 100000 140000 20000 ...
## $ Male : int 0 0 0 0 1 1 1 0 0 1 ...
## $ Education : int 2 2 2 2 2 1 1 2 3 3 ...
## $ MaritalStatus : int 1 2 2 1 1 2 2 2 1 2 ...
## $ Age : int 24 26 34 37 57 37 29 23 28 35 ...
## $ BillOutstanding: int 3913 2682 29239 46990 8617 64400 367965 11876 11285 0 ...
## $ LastPayment : int 0 0 1518 2000 2000 2500 55000 380 3329 0 ...
## $ Default : int 1 1 0 0 0 0 0 0 0 0 ...
## - attr(*, ".internal.selfref")=<externalptr># convert 'Id' as a factor
#CCdefault.dt[, Id := as.factor(Id)]
# convert 'Male' as a factor
CCdefault.dt[, Male := as.factor(Male)]
# convert 'Education' as a factor
CCdefault.dt[, Education := as.factor(Education)]
# convert 'MaritalStatus' as a factor
CCdefault.dt[, MaritalStatus := as.factor(MaritalStatus)]
# convert 'Default' as a factor
CCdefault.dt[, Default := as.factor(Default)]
# Changing the lavels of 'Default' variable
levels(CCdefault.dt$Default) <- c("No","Yes")
# verifying conversion
str(CCdefault.dt)## Classes 'data.table' and 'data.frame': 29601 obs. of 9 variables:
## $ Id : int 1 2 3 4 5 6 7 8 9 10 ...
## $ CreditLimit : int 20000 120000 90000 50000 50000 50000 500000 100000 140000 20000 ...
## $ Male : Factor w/ 2 levels "0","1": 1 1 1 1 2 2 2 1 1 2 ...
## $ Education : Factor w/ 4 levels "1","2","3","4": 2 2 2 2 2 1 1 2 3 3 ...
## $ MaritalStatus : Factor w/ 3 levels "1","2","3": 1 2 2 1 1 2 2 2 1 2 ...
## $ Age : int 24 26 34 37 57 37 29 23 28 35 ...
## $ BillOutstanding: int 3913 2682 29239 46990 8617 64400 367965 11876 11285 0 ...
## $ LastPayment : int 0 0 1518 2000 2000 2500 55000 380 3329 0 ...
## $ Default : Factor w/ 2 levels "No","Yes": 2 2 1 1 1 1 1 1 1 1 ...
## - attr(*, ".internal.selfref")=<externalptr>#PART 3: Check / Reset the Levels of the target variable
# levels of the target variable
levels(CCdefault.dt$Default)## [1] "No" "Yes"# ordering the levels
CCdefault.dt$Default <- ordered(CCdefault.dt$Default, levels = c("Yes", "No"))
# verifying the new order of levels
levels(CCdefault.dt$Default)## [1] "Yes" "No"#PART 4: Split the data into a training set (80%) and a testing set (20%)
library(caret)## Loading required package: lattice## Loading required package: ggplot2## Registered S3 methods overwritten by 'ggplot2':
## method from
## [.quosures rlang
## c.quosures rlang
## print.quosures rlang# data partition
set.seed(2341)
trainIndex <- createDataPartition(CCdefault.dt$Default, p = 0.80, list = FALSE)
# 80% training data
trainData.dt <- CCdefault.dt[trainIndex, ]
# 20% testing data
testData.dt <- CCdefault.dt[-trainIndex, ]#PART 5: Verify the Split
# dimension of training dataset
dim(trainData.dt)## [1] 23681 9# dimension of testing dataset
dim(testData.dt)## [1] 5920 9# proportion of defaulters in training dataset
round(prop.table(table(trainData.dt$Default))*100,2)##
## Yes No
## 22.31 77.69# proportion of defaulters in test dataset
round(prop.table(table(testData.dt$Default))*100,2)##
## Yes No
## 22.31 77.69#Run the Machine Learning algorithm – kNN
library(caret)
# Set control parameters
trctrl <- trainControl(method = "repeatedcv",
number = 10,
repeats = 3)
set.seed(3333)
# Run kNN Classifier in package caret
knn_fit <- train(Default ~ CreditLimit + Male + Education + MaritalStatus + Age+ BillOutstanding + LastPayment,
data = trainData.dt,
method = "knn",
trControl = trctrl,
preProcess = c("center", "scale"),
tuneLength = 10)
# kNN model summary
knn_fit ## k-Nearest Neighbors
##
## 23681 samples
## 7 predictor
## 2 classes: 'Yes', 'No'
##
## Pre-processing: centered (10), scaled (10)
## Resampling: Cross-Validated (10 fold, repeated 3 times)
## Summary of sample sizes: 21313, 21314, 21313, 21313, 21312, 21314, ...
## Resampling results across tuning parameters:
##
## k Accuracy Kappa
## 5 0.7363992 0.05716480
## 7 0.7483638 0.05526080
## 9 0.7558805 0.05114314
## 11 0.7623696 0.04975468
## 13 0.7649173 0.04025378
## 15 0.7666202 0.03245795
## 17 0.7684078 0.02854856
## 19 0.7693933 0.02367882
## 21 0.7711811 0.02270309
## 23 0.7725323 0.02163401
##
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was k = 23.# predicting the test set observations
kNNPred <- predict(knn_fit, testData.dt, type = "raw")
# confusion matrix
table(Predicted = kNNPred, Actual = testData.dt$Default)## Actual
## Predicted Yes No
## Yes 24 58
## No 1297 4541# loading the package
library(ROCR)## Loading required package: gplots##
## Attaching package: 'gplots'## The following object is masked from 'package:stats':
##
## lowesskNNPred <- predict(knn_fit, testData.dt,type = "raw")
kNNPredObj <- prediction(as.numeric(kNNPred),as.numeric(testData.dt$Default))
kNNPerfObj <- performance(kNNPredObj, "tpr","fpr")
# plotting ROC curve
plot(kNNPerfObj,main = "ROC Curve",col = 2,lwd = 2)
abline(a = 0,b = 1,lwd = 2,lty = 3,col = "black")
# area under curve
aucLR <- performance(kNNPredObj, measure = "auc")
aucLR <- aucLR@y.values[[1]]
aucLR## [1] 0.5027783# trainData.dt
# testData.dt