---
title: "Business Data Analytics. Practice Session"
subtitle: "Customer Lifecycle Management: Churn Prediction"
output:
  prettydoc::html_pretty: null
  highlight: github
  html_document: default
  html_notebook: default
  github_document: default
  theme: cayman
---


##Step 1: Data loading
```{r}


library('tidyverse')
library('data.table')

# Load the dataset

dt <- read.csv(file.choose())
dt <- as.data.frame(dt)

# check the dataset
str(dt)

#check the data type of each column
sapply(dt, class)

```

##Step 2: Data wrangling and transformation
```{r}
# check the columns with NA values
sapply(dt, function(x) sum(is.na(x)))

# remove the rows with NA values
dt <- dt[complete.cases(dt) == TRUE, ]


# check the unique calues for each column

for (col in names(dt)){
    if (class(dt[[col]]) %in% c("character", "factor") & col != 'customerID'){
          print(unique(dt[col]))  
    }
}

#need to change "No internet service" to "No" for "OnlineSecurity", #"OnlineBackup", "DeviceProtection", "TechSupport", "streamingTV", "streamingMovies".

for(i in c(10:15)) {
  print(names(dt[i]))
}

for(i in c(10:15)) {
  dt[i][dt[i]=="No internet service"] <- "No"
}

# change "No phone service" to "No" for MultipleLines
dt$MultipleLines[dt$MultipleLines == "No phone service"] <- "No"


#remove CustomerID column
dt$customerID <- NULL

#check tenure
min(dt$tenure); max(dt$tenure)

#transform tenure to 
dt$tenure <- as.factor(ifelse(dt$tenure == 0, '0',
                   ifelse(dt$tenure <= 24, '1',
                          ifelse(dt$tenure <= 48, '2', '3'
                          ))))

#tranform Churn values from Yes-no to 1-0
dt$Churn <- as.factor(ifelse(dt$Churn == "Yes", 1, 0))

#change SeniorCitizen type to factor
dt$SeniorCitizen <- as.factor(dt$SeniorCitizen)

# transform the needed columns to factor data type

for (col in names(dt)){
    if (class(dt[[col]]) == "character"){
          dt <- dt%>%
          mutate_at(col, funs(factor(.)))
    }
}


#check the data type of each column again
str(dt)

```

## Step 3: Preparing test and training data
```{r}

#Check the numeric variables and correlation between them
cols <- sapply(dt, is.numeric)


cor(dt%>%select(MonthlyCharges, TotalCharges))


dt$TotalCharges <- NULL

#generate random sample 
set.seed(999) 
sample <- sample.int(n = nrow(dt), size = floor(0.7*nrow(dt)), replace = F)

# split data to train and test datasets
train <- dt[sample, ]
test  <- dt[-sample, ]

dim(train); dim(test)



``` 

##First Model: General linear model

```{r}

#create logistic regression.
LogModel <- glm(Churn ~ .,family=binomial(link="logit"),data=train)
print(summary(LogModel))

#analyze the deviance table with anova
anova(LogModel, test="Chisq")

#create prediction model
predict_glm <- predict(LogModel, newdata = test[, -19], type = 'response')


test$Churn <- as.character(test$Churn)
test$Churn[test$Churn=="No"] <- "0"
test$Churn[test$Churn=="Yes"] <- "1"


#set the predicted value of churn by threshold = 0.5
predict_glm <- ifelse(predict_glm >0.5, 1, 0)

# get the accuracy of the model
misClasificError <- mean(predict_glm != test$Churn)
print(paste('Logistic Regression Accuracy',1-misClasificError))

# Confusion matrix
table(test$Churn, predict_glm)

```

## Second Model: Decision Tree
```{r}
#way 1: with library "party"
library("party")

tree <- ctree(Churn~Contract+tenure+PaperlessBilling, train)
plot(tree, type='simple')


# Make prediction on Test dataset
pred_tree <- predict(tree, test)

#Confusion Matrix for Decision Tree
table(Predicted = pred_tree, Actual = test$Churn)


p1 <- predict(tree, train)
tab1 <- table(Predicted = p1, Actual = train$Churn)
tab2 <- table(Predicted = pred_tree, Actual = test$Churn)

#Accuracy of Decision tree model on Train dataset
(tab1[1,1] + tab1[2,2]) / sum(tab1)

#Accuracy of Decision tree model on Test dataset
(tab2[1,1] + tab2[2,2]) / sum(tab2)


# way2: with library "rpart"
library(rpart)
library("rpart.plot")
tree2 <- rpart(data=train, Churn ~ Contract+tenure+PaperlessBilling, method='class')

rpart.plot(tree2)

```

## Third Model: Random Forrest
```{r}
library("randomForest")

#create model with training data and see the confusion matrix 
rfModel <- randomForest(Churn ~., data = train)
print(rfModel)


# predict churn on the test dataset with the created model 
pred_rf <- predict(rfModel, test)

# check confusion matrix
table(pred_rf, as.factor(test$Churn))
# plot model to check error dynamics for the different number of trees
plot(rfModel)

varImpPlot(rfModel, sort=T, n.var = 10, main = 'Top 10 Feature Importance')


```