baza<-read.csv("C:\\Users\\Aleksandar\\Desktop\\projekat-biostat\\abcChurn.csv")

#izbacivanje kolona X i ID jer su nepotrebne u
#odredjivanju promenljive churn
baza<-baza[,-c(1,2)]

summary(baza)
str(baza)

levels(baza$grad)
baza$grad<- as.numeric(baza$grad)
str(baza$pol)
baza$pol<- as.numeric(baza$pol)
baza$pol[baza$pol==""]<-NA
length(which(is.na(baza$pol)))
#NA vrednosti prebacujemo u one kojih je vise, dakle u male
xtabs(~pol, data=baza)
baza$pol[is.na(baza$pol)]<-"male"
library(corrplot)
#Pravi se korelaciona matrica da se proveri
#sa kojom vrednoscu je churn u jakoj zavisnosti
corr.matrix<-cor(baza)
corrplot(corr.matrix, method="number", type="upper", diag=F, number.cex = 0.5, tl.cex=0.5)

baza1<-baza[,c("cena_izabranog_plana","uplaceno")]

summary(baza1)
vars<- c("cena_izabranog_plana","uplaceno")
library(dplyr)
set.seed(23)
data.st<- baza1[, vars]
data.st <- baza %>% mutate_each_(funs(scale(.) %>% as.vector),
                                  vars=c("cena_izabranog_plana", "uplaceno","datumIsticanjaClanstva"))
churn.st<- data.st[, vars]

churn.st<- as.data.frame(churn.st)
churn.st<-as.factor(baza$churn)
churn.st$churn<-as.factor(baza$churn)
summary(churn.st)
boxplot(churn.st)

#Sad radimo drvo odlucivanja
library(caret)
?caret
set.seed(11)
train.indices <- createDataPartition(baza$churn,p=0.8,list=FALSE)
train.data <- churn.st[train.indices,]
test.data <- churn.st[-train.indices,]
prop.table(table(train.data$churn))
prop.table(table(test.data$churn))
library(e1071)
library(rpart)
library(rpart.plot)
numFolds = trainControl( method = "cv", number = 10 )
cpGrid = expand.grid( .cp = seq(0.001, to = 0.05, by = 0.001))
set.seed(10)
dt.cv <- train(churn ~  cena_izabranog_plana + uplaceno, data = train.data, method = "rpart", control = rpart.control(minsplit = 10), trControl = numFolds, tuneGrid = cpGrid)
#cp=0.001
tree<-rpart(churn ~.,data=train.data, method = "class",
            control=rpart.control(minsplit = 10,cp=0.001))
library(rpart.plot)
prp(tree)
tree.pred <- predict(tree, newdata = test.data, type = "class")
tree.cm <- table(true = test.data$churn, predicted = tree.pred)
tree.cm
compute.eval.metrics <- function(cmatrix) {
  TP <- cmatrix[1,1]
  TN <- cmatrix[2,2]
  FP <- cmatrix[2,1]
  FN <- cmatrix[1,2]
  acc = sum(diag(cmatrix)) / sum(cmatrix)
  precision <- TP / (TP + FP)
  recall <- TP / (TP + FN)
  F1 <- 2*precision*recall / (precision + recall)
  c(accuracy = acc, precision = precision, recall = recall, F1 = F1)
}
tree.eval <- compute.eval.metrics(tree.cm)
tree.eval