credit1<-read.table("credit.csv",header=T,sep=",")

dim(credit1)
summary(credit1)

credit1$Cardhldr<-as.factor(credit1$Cardhldr)
credit1$Ownrent<-as.factor(credit1$Ownrent)
credit1$Selfempl<-as.factor(credit1$Selfempl)

summary(credit1)

library(MASS)
lda1<-lda(Cardhldr~., data=credit1)
lda1
summary(lda1)

lda1.pred<-predict(lda1,credit1)$class
mean(ifelse(lda1.pred!=credit1$Cardhldr,1,0))
table(lda1.pred,credit1$Cardhldr)


logis1<-glm(Cardhldr~., data=credit1,family=binomial)
summary(logis1)
logis1.pred<-predict(logis1,credit1,"response")
mean(ifelse(logis1.pred!=credit1$Cardhldr,1,0))
table(logis1.pred,credit1$Cardhldr)

library(nnet)
logis2<-multinom(Cardhldr~., data=credit1)
logis2

logis2.pred<-predict(logis2,credit1)
mean(ifelse(logis2.pred!=credit1$Cardhldr,1,0))
table(logis2.pred,credit1$Cardhldr)


logis3<-step(logis2)
#summary(logis2)
logis3.pred<-predict(logis3,credit1)
mean(ifelse(logis3.pred!=credit1$Cardhldr,1,0))
table(logis3.pred,credit1$Cardhldr)
#check the coeff


library(tree)
tr1<-tree(Cardhldr~., data=credit1)
plot(tr1);text(tr1)
## plot looks funny, what is Avgexp? why it can classifiy so well?
tr1.pred<-predict(tr1,credit1,"class")
mean(ifelse(tr1.pred!=credit1$Cardhldr,1,0))
table(tr1.pred,credit1$Cardhldr)

library(randomForest)
rf1<-randomForest(Cardhldr~., data=credit1)
plot(rf1)
rf1.pred<-predict(rf1,credit1)
mean(ifelse(rf1.pred!=credit1$Cardhldr,1,0))
table(rf1.pred,credit1$Cardhldr)
## don't forget RF performance in training set is sometims too good!!

library(nnet)
nn1<-nnet(Cardhldr~., data=credit1, size=10, decay=0.1)
nn1.pred<-predict(nn1,credit1,"class")
mean(ifelse(nn1.pred!=credit1$Cardhldr,1,0))
table(nn1.pred,credit1$Cardhldr)

library(e1071)
svm1<-svm(Cardhldr~., data=credit1)
svm1.pred<-predict(svm1,credit1)
mean(ifelse(svm1.pred!=credit1$Cardhldr,1,0))
table(svm1.pred,credit1$Cardhldr)


##now train/test computation

p<-0.7   # proportion of training set
n<-nrow(credit1)


rep.n<-10  # number of repetition

misclass<-matrix(rep(0,rep.n*6),ncol=6)
for (i in 1:rep.n){
        train.ind<-sample(n,as.integer(n*p))  #get train index randomly
	   train.data<-credit1[train.ind,]       #get train data
        test.data<-credit1[-train.ind,]       #get test data

	lda1<-lda(Cardhldr~., data=train.data)
	lda1.pred<-predict(lda1,test.data)$class
	
	logis1<-multinom(Cardhldr~., data=train.data)
	logis1.pred<-predict(logis1,test.data)

	tr1<-tree(Cardhldr~., data=train.data)
	tr1.pred<-predict(tr1,test.data,"class")
	

	nn1<-nnet(Cardhldr~., data=train.data,size=10)
	nn1.pred<-predict(nn1,test.data,"class")
	
	svm1<-svm(Cardhldr~., data=train.data)
	svm1.pred<-predict(svm1,test.data)

	rf1<-randomForest(Cardhldr~., data=train.data)
	rf1.pred<-predict(rf1,test.data)

	misclass[i,1]<-mean(ifelse(test.data$Cardhldr==lda1.pred,0,1))
	misclass[i,2]<-mean(ifelse(test.data$Cardhldr==logis1.pred,0,1))
	misclass[i,3]<-mean(ifelse(test.data$Cardhldr==tr1.pred,0,1))
	misclass[i,4]<-mean(ifelse(test.data$Cardhldr==nn1.pred,0,1))
	misclass[i,5]<-mean(ifelse(test.data$Cardhldr==svm1.pred,0,1))
	misclass[i,6]<-mean(ifelse(test.data$Cardhldr==rf1.pred,0,1))
	print(i)
}

colMeans(misclass)
boxplot(misclass)

## what if we don't have Exp_inc and Avgexp

credit2<-credit1[,-c(5,6)]


p<-0.7   # proportion of training set
n<-nrow(credit2)


rep.n<-10  # number of repetition

misclass<-matrix(rep(0,rep.n*6),ncol=6)
for (i in 1:rep.n){
        train.ind<-sample(n,as.integer(n*p))  #get train index randomly
	   train.data<-credit2[train.ind,]       #get train data
        test.data<-credit2[-train.ind,]       #get test data

	lda1<-lda(Cardhldr~., data=train.data)
	lda1.pred<-predict(lda1,test.data)$class
	
	logis1<-multinom(Cardhldr~., data=train.data)
	logis1.pred<-predict(logis1,test.data)

	tr1<-tree(Cardhldr~., data=train.data)
	tr1.pred<-predict(tr1,test.data,"class")
	

	nn1<-nnet(Cardhldr~., data=train.data,size=10)
	nn1.pred<-predict(nn1,test.data,"class")
	
	svm1<-svm(Cardhldr~., data=train.data)
	svm1.pred<-predict(svm1,test.data)

	rf1<-randomForest(Cardhldr~., data=train.data)
	rf1.pred<-predict(rf1,test.data)

	misclass[i,1]<-mean(ifelse(test.data$Cardhldr==lda1.pred,0,1))
	misclass[i,2]<-mean(ifelse(test.data$Cardhldr==logis1.pred,0,1))
	misclass[i,3]<-mean(ifelse(test.data$Cardhldr==tr1.pred,0,1))
	misclass[i,4]<-mean(ifelse(test.data$Cardhldr==nn1.pred,0,1))
	misclass[i,5]<-mean(ifelse(test.data$Cardhldr==svm1.pred,0,1))
	misclass[i,6]<-mean(ifelse(test.data$Cardhldr==rf1.pred,0,1))
	print(i)
}

colMeans(misclass)
boxplot(misclass)
colnames(misclass)<-c("lda","logistic","tree","nnet","svm","RF")
boxplot(misclass)




colMeans(misclass)
boxplot(misclass)