ClassPractice9.R

# Set up working directory
	

	setwd("C:\\Users\\Vatsal\\Desktop\\AMMA\\data_2017")
	

	# Read data
	card_balance<- read.csv(file = "card_balance.csv", stringsAsFactors = F)
	

	

	# --------------------- Multiple Regression -----------------------------
	

	sm_f <- sample(1:nrow(card_balance),0.6*nrow(card_balance), replace = F)
	sm_card_balance <- card_balance[sm_f,]
	

	

	# Obj: Estimate billed amount for next 3 months
	

	names(sm_card_balance)
	

	summary(sm_card_balance)
	

	# We can remove the customers who do not have any billed amount in the first 3 months
	sm_card_balance$Pre_billed_amt_3m <-apply(sm_card_balance[,c(7,8,9)],1, sum)
	

	summary(sm_card_balance)
	quantile(sm_card_balance$Pre_billed_amt_3m, probs = seq(0,1,0.05))
	

	

	# Exclude with negative billed amount
	sm_card_balance <- sm_card_balance[sm_card_balance$Pre_billed_amt_3m>30,]
	

	# Target Variable: Average Balance in next 3 month
	sm_card_balance$billed_3m <-apply(sm_card_balance[,c(10,11,12)],1, mean)
	

	

	quantile(sm_card_balance$billed_3m, probs = seq(0,1,0.05))
	

	hist(sm_card_balance$billed_3m,breaks = 20)
	

	quantile(sm_card_balance$billed_3m, probs = seq(0,1,0.05))
	

	# Outlier Treatment of Target Variables
	sm_card_balance$billed_3m <- ifelse(sm_card_balance$billed_3m>165000,165000,
	                                 ifelse(sm_card_balance$billed_3m<0,0,sm_card_balance$billed_3m))
	

	names(sm_card_balance)
	

	# Regression MOdel fitting
	card_billed_amt_reg <- lm(billed_3m~AGE+Pre_billed_amt_3m+Billed_amount_1+Billed_amount_2,
	                data=sm_card_balance)
	

	summary(card_billed_amt_reg)
	

	sm_card_balance$pred_bill_amt <- predict(card_billed_amt_reg,sm_card_balance)
	

	plot(sm_card_balance$billed_3m,
	     sm_card_balance$pred_bill_amt,
	     pch=20,
	     col="red")
	abline(lm(sm_card_balance$billed_3m ~ sm_card_balance$pred_bill_amt),col="blue",lwd=4)
	

	# --------------------- Logistic Regression -----------------------------
	

	# Target Variable: If Billed Amount reduces below 20%
	

	names(sm_card_balance)
	

	sm_card_balance$Pred_bill_drop <- ifelse (sm_card_balance$billed_3m < 0.2*sm_card_balance$Pre_billed_amt_3m,1,0)
	

	table(sm_card_balance$Pred_bill_drop)
	

	# Demographic Variables
	

	Billed_amt_drop_logit <- glm(formula = Pred_bill_drop~
	                               AGE+Pre_billed_amt_3m+Billed_amount_1+Payment_1+Payment_2+Payment_3+
	                               Billed_amount_2
	                             ,
	                      family=binomial,
	                      data=sm_card_balance)
	summary(Billed_amt_drop_logit)
	

	# Predict
	sm_card_balance$Pred_Amt_drop_prob <- predict(Billed_amt_drop_logit,
	                                              sm_card_balance,
	                              type = c("response"))
	

	table(sm_card_balance$Pred_bill_drop)/nrow(sm_card_balance)
	quantile(sm_card_balance$Pred_Amt_drop_prob, 
	         probs = seq(0,1,0.05))
	

	sm_card_balance$Pred_Amt_drop_Class <- ifelse(sm_card_balance$Pred_Amt_drop_prob >0.288,1,0)
	

	table(sm_card_balance$Pred_Amt_drop_Class,sm_card_balance$Pred_bill_drop)
	

	library(caret)
	

	# Create Confusion Matrix
	confusionMatrix(data=factor(sm_card_balance$Pred_Amt_drop_Class),
	                reference=sm_card_balance$Pred_bill_drop,
	                positive='1')
	

	library("ROCR")
	

	perf.obj <- prediction(predictions=sm_card_balance$Pred_Amt_drop_Class,
	                       labels=sm_card_balance$Pred_bill_drop)
	# Get data for ROC curve
	roc.obj <- performance(perf.obj, measure="tpr", x.measure="fpr")
	plot(roc.obj,
	     main="Balance Amt Drop - ROC Curves",
	     xlab="1 - Specificity: False Positive Rate",
	     ylab="Sensitivity: True Positive Rate",
	     col="blue")
	abline(0,1,col="grey")
	

	#--------------------------  Decision Tree: CART -----------------------------------
	

	install.packages("rpart")
	install.packages("rpart.plot")
	

	library(rpart)
	dt1 <- rpart(Pred_bill_drop~AGE+Pre_billed_amt_3m+Billed_amount_1+Payment_1+Payment_2+Payment_3+
	               Billed_amount_2,
	             data=sm_card_balance)
	

	library(rpart.plot)
	rpart.plot(dt1)
	

	install.packages("libcoin")
	install.packages("partykit")
	install.packages("CHAID", repos="http://R-Forge.R-project.org")
	library(CHAID)
	library(help=CHAID)
	

	sm_card_balance1 <- sm_card_balance
	sm_card_balance1$AGE_cat <- cut(sm_card_balance1$AGE,
	                                quantile(sm_card_balance1$AGE, probs = seq(0,1,0.05)))
	sm_card_balance1$Pre_billed_amt_3m_cat <- cut(sm_card_balance1$Pre_billed_amt_3m,
	                                quantile(sm_card_balance1$Pre_billed_amt_3m, probs = seq(0,1,0.05)))
	sm_card_balance1$Payment_1_cat <- cut(sm_card_balance1$Payment_1,
	                                      unique(quantile(sm_card_balance1$Payment_1, probs = seq(0,1,0.05))))
	

	dt.chaid <- chaid(as.factor(Pred_bill_drop)~AGE_cat+Pre_billed_amt_3m_cat+Payment_1_cat,
	             data=sm_card_balance1)
	

	

	dt.chaid
	

	plot(dt.chaid)