MATH 564 - Final Project.Rmd

---
title: "Applied Statistics Project"
author: "Amogh" "Rahul" "Akshay"
date: "2022-10-17"
output: pdf_document
---

```{r cars}
library(plyr)
library(corrplot)
library(ggplot2)
library(gridExtra)
library(ggthemes)
library(caret)
library(lattice)
library(MASS)
library(randomForest)
library(party)
library(sandwich)
library(rpart)
library(rattle)
library(GoodmanKruskal)
library(e1071)
library(rpart.plot)
library(caTools)
library(class)
```

```{r}
churn <- read.csv('C:\\Users\\koria\\Downloads\\BankChurners.csv')
#View(churn)
str(churn)
```

```{r}
sapply(churn, function(x) sum(is.na(x)))

churn$Naive_Bayes_Classifier_Attrition_Flag_Card_Category_Contacts_Count_12_mon_Dependent_count_Education_Level_Months_Inactive_12_mon_1  <- NULL
churn$Naive_Bayes_Classifier_Attrition_Flag_Card_Category_Contacts_Count_12_mon_Dependent_count_Education_Level_Months_Inactive_12_mon_2 <- NULL
churn$CLIENTNUM <- NULL
View(churn)
```

```{r}
str(churn)
summary(churn)
```

```{r}
print(head(churn[1:3,]))
```

```{r}
#Data Exploration
ggplot(churn, aes(x=Gender)) + 
  geom_bar(stat="count") + 
  labs(title= "Attrition number by Gender", x= "Gender", y="Attrition") + 
  theme_classic() + scale_color_brewer(palette="Set2")

ggplot(churn, aes(x=Customer_Age)) + 
  geom_bar(stat="count") + 
  labs(title= "Attrition number by Age", x= "Age", y="Attrition") + 
  theme_classic() + scale_color_brewer(palette="Set2")

ggplot(churn, aes(x=Education_Level)) + 
  geom_bar(stat="count") + 
  labs(title= "Attrition number by Education", x= "Education", y="Attrition") + 
  theme_classic() + scale_color_brewer(palette="Set2")

ggplot(churn, aes(x=Income_Category)) + 
  geom_bar(stat="count") + 
  labs(title= "Attrition number by Income", x= "Income", y="Attrition") + 
  theme_classic() + scale_color_brewer(palette="Set2")

ggplot(churn, aes(x=Card_Category)) + 
  geom_bar(stat="count") + 
  labs(title= "Attrition number by Card Type", x= "Card Type", y="Attrition") + 
  theme_classic() + scale_color_brewer(palette="Set2")

```

```{r}
table(churn$Attrition_Flag, churn$Customer_Age)
table(churn$Attrition_Flag, churn$Gender)

```

```{r}
ggplot(churn, aes(x=Attrition_Flag,
                  y= prop.table(stat(count)),
                  fill= factor(Gender),
                  label= scales::percent(prop.table(stat(count))))) +
  geom_bar(position = position_dodge())+
  geom_text(stat="count",
            position = position_dodge(.9),
            vjust= -0.5, size=3)+
  scale_y_continuous(labels = scales::percent)+
  labs(title = "Attrition by Gender",
       x= "Attrition status",
       y="Count")+
  theme_classic()+
  scale_fill_discrete(
    name="Gender",
    breaks=c("M", "F"),
    labels=c("Male", "Female" )
  )

ggplot(churn, aes(x=Attrition_Flag,
                  y= prop.table(stat(count)),
                  fill= factor(Card_Category),
                  label= scales::percent(prop.table(stat(count))))) +
  geom_bar(position = position_dodge())+
  geom_text(stat="count",
            position = position_dodge(.9),
            vjust= -0.5, size=3)+
  scale_y_continuous(labels = scales::percent)+
  labs(title = "Attrition by Card Category",
       x= "Attrition status",
       y="Count")+
  theme_classic()

ggplot(churn, aes(x=Attrition_Flag,
                  y= prop.table(stat(count)),
                  fill= factor(Income_Category),
                  label= scales::percent(prop.table(stat(count))))) +
  geom_bar(position = position_dodge())+
  geom_text(stat="count",
            position = position_dodge(.9),
            vjust= -0.5, size=3)+
  scale_y_continuous(labels = scales::percent)+
  labs(title = "Attrition by Income Category",
       x= "Attrition status",
       y="Count")+
  theme_classic()

ggplot(churn, aes(y=Customer_Age,
                  x= Education_Level,
                  fill= factor(Attrition_Flag))) +
  geom_boxplot(position = position_dodge())+
  labs(title = "Attrition Status By Age and Education",
       x= "Education level",
       y="Age")+ theme_light()

ggplot(churn, aes(Months_on_book))+
  geom_density(col="blue")+ facet_wrap(~Attrition_Flag)+theme_bw()

ggplot(churn, aes(Marital_Status))+
  geom_density(col="blue")+ facet_wrap(~Attrition_Flag)+theme_bw()

ggplot(churn, aes(Dependent_count))+
  geom_density(col="blue")+ facet_wrap(~Attrition_Flag)+theme_bw()
```

```{r}
#PCA starts here
#PCA
churn.pca <- prcomp(scale(churn[,c(2,4,9:20)]), center = TRUE)
summary(churn.pca)
```

```{r}
pc_data <- churn.pca$x[,1:10]
cat_data <- churn[,c(1,3,5:8)]
churn_pca <-data.frame(cat_data, pc_data)
```

```{r}
churn_pca[sapply(churn_pca, is.character)]<- lapply(churn_pca[sapply(churn_pca, is.character)], as.factor)
summary(churn_pca)
```

```{r}
#Plotting PCA
plot(churn.pca$x[,1],churn.pca$x[,2])
```

```{r}
#How much variation in the original data does PCA account for
churn.pca.var <- churn.pca$sdev^2
churn.pca.var.per <- round(churn.pca.var/sum(churn.pca.var)*100,1)
churn.pca.var.per
```

```{r}
#Plotting PCA percentages
barplot(churn.pca.var.per, main="Scree Plot", xlab="Principal Component Analysis", names = c("PC1", "PC2", "PC3", "PC4", "PC5", "PC6", "PC7", "PC8", "PC9", "PC10", "PC11", "PC12", "PC13","PC14"), ylab="Percent Variation", )
#PCA ends here
```

```{r}
#Converting all features to categorical data
churn[sapply(churn, is.character)]<- lapply(churn[sapply(churn, is.character)], as.factor)
```

```{r}
str(churn)
summary(churn)
```

**************************************************************************************************************
#Dot Charts

```{r}
library(leaps)
library(olsrr)
library(car)
library(corrplot)
View(churn_pca)
```

```{r}
par(mfrow = c(2 , 2))
dotchart(churn_pca$PC1, labels = 'PC1')
dotchart(churn_pca$PC2, labels = 'PC2')
dotchart(churn_pca$PC3, labels = 'PC3')
dotchart(churn_pca$PC4, labels = 'PC4')
dotchart(churn_pca$PC5, labels = 'PC5')
dotchart(churn_pca$PC6, labels = 'PC6')
dotchart(churn_pca$PC7, labels = 'PC7')
dotchart(churn_pca$PC8, labels = 'PC8')
dotchart(churn_pca$PC9, labels = 'PC9')
dotchart(churn_pca$PC10, labels = 'PC10')

```

```{r}
plot(churn_pca[,7:16])
```

```{r}
correlation <- cor(churn_pca[,7:16])
corrplot(correlation,method = 'number')

```

```{r}
#Splitting the pca dataset
intrain_pca<- createDataPartition(churn_pca$Attrition_Flag, p=0.80, list = FALSE)
training_pca<- churn_pca[intrain_pca,]
testing_pca<- churn_pca[-intrain_pca,]
dim(training_pca); dim(testing_pca)
#summary(training_pca)
#summary(testing_pca)
```

```{r}
#Splitting the regular dataset
intrain_reg<- createDataPartition(churn$Attrition_Flag, p=0.80, list = FALSE)
training_reg<- churn[intrain_reg,]
testing_reg<- churn[-intrain_reg,]
dim(training_reg); dim(testing_reg)
#summary(training_reg)
#summary(testing_reg)
```

```{r}
#Randomforest for PCA data
random_forest <- randomForest(Attrition_Flag ~ ., ntree= 500, family="binomial", data=training_pca)
print(summary(random_forest))
random_forest
rf_pred <- predict(random_forest, testing_pca)
caret::confusionMatrix(rf_pred, testing_pca$Attrition_Flag)
```

```{r}
#Randomforest for Regular data
random_forest <- randomForest(Attrition_Flag ~ ., ntree= 500, family="binomial", data=training_reg)
print(summary(random_forest))
random_forest
rf_pred <- predict(random_forest, testing_reg)
caret::confusionMatrix(rf_pred, testing_reg$Attrition_Flag)
```

```{r}
#Logistic Regression for PCA Data
LogModel <- glm(Attrition_Flag ~ ., family= "binomial", data = training_pca)
print(summary(LogModel))
anova(LogModel, test="Chisq")
log_reg <- predict(LogModel, testing_pca[-1],  type = "response")
y_pred <- ifelse(log_reg > 0.5, 2, 1)
y_pred <- as.numeric(y_pred)
target <- as.numeric(testing_pca$Attrition_Flag)
#prop.table(table(training_pca$Attrition_Flag))
caret::confusionMatrix(table(y_pred, target)) 

```

```{r}
#Logistic Regression for Regular Data
LogModel <- glm(Attrition_Flag ~ ., family= "binomial", data = training_reg)
print(summary(LogModel))
anova(LogModel, test="Chisq")
log_reg <- predict(LogModel, testing_reg[-1],  type = "response")
y_pred <- ifelse(log_reg > 0.5, 2, 1)
y_pred <- as.numeric(y_pred)
target <- as.numeric(testing_pca$Attrition_Flag)
#prop.table(table(training_pca$Attrition_Flag))
caret::confusionMatrix(table(y_pred, target)) 

```

```{r}
#SVM for PCA Data
svmfit = svm(Attrition_Flag ~ ., data = training_pca, cross = 10, gamma = 0.5, cost = 1)
svm_pred <- predict(svmfit, testing_pca)
summary(svmfit)
caret::confusionMatrix(svm_pred, testing_pca$Attrition_Flag)
```

```{r}
#SVM for Regular Data
svmfit = svm(Attrition_Flag ~ ., data = training_reg, cross = 10, gamma = 0.5, cost = 1)
svm_pred <- predict(svmfit, testing_reg)
summary(svmfit)
caret::confusionMatrix(svm_pred, testing_reg$Attrition_Flag)
```

```{r}
#Naive Bayes for PCA Data
naive_bayes<- naiveBayes(Attrition_Flag ~ ., data= training_pca)
naive_bayes
nb_pred<- predict(naive_bayes, testing_pca)
caret::confusionMatrix(nb_pred, testing_pca$Attrition_Flag)
```

```{r}
#Naive Bayes for Regular Data
naive_bayes<- naiveBayes(Attrition_Flag ~ ., data= training_reg)
naive_bayes
nb_pred<- predict(naive_bayes, testing_reg)
caret::confusionMatrix(nb_pred, testing_reg$Attrition_Flag)
```

```{r}
#Decision tree for PCA data
decision_tree <- ctree(Attrition_Flag ~ ., data= training_pca)
decision_tree
dt_pred<- predict(decision_tree, testing_pca)
caret::confusionMatrix(dt_pred, testing_pca$Attrition_Flag)
```

```{r}
#Decision tree for Regular data
decision_tree <- ctree(Attrition_Flag ~ ., data= training_reg)
decision_tree
dt_pred<- predict(decision_tree, testing_reg)
caret::confusionMatrix(dt_pred, testing_reg$Attrition_Flag)
```

```{r}
# Comparision of different models on PCA Data 

H = c(91.26,87.21,89.78,88.74,90.07)
names1 = c("Random Forest","SVM" , "Naive Bayes","Decision Tree","Logistic Regression")
experiment <- data.frame(Algorithm = names1,
                         Percentage = H)
ggplot(data = experiment, mapping = aes(x=Algorithm, y=Percentage)) +
  geom_bar(stat="identity", position = "dodge",fill="lightblue") + scale_fill_brewer(palette = "Pastel2")+
  geom_text(aes(label = Percentage), vjust = -0.2, size = 5,
            position = position_dodge(0.9)) +
  ylim(0, max(experiment$Percentage)*1.1)
```

```{r}
# Comparision of different models on Regular Data 

H = c(96.35,86.86,89.33,94.07, 76.05)
names1 = c("Random Forest","SVM" , "Naive Bayes","Decision Tree","Logistic Regression")
experiment <- data.frame(Algorithm = names1,
                         Percentage = H)
ggplot(data = experiment, mapping = aes(x=Algorithm, y=Percentage)) +
  geom_bar(stat="identity", position = "dodge",fill="lightblue") + scale_fill_brewer(palette = "Pastel2")+
  geom_text(aes(label = Percentage), vjust = -0.2, size = 5,
            position = position_dodge(0.9)) +
  ylim(0, max(experiment$Percentage)*1.1)
```

```{r}
# COMPARISION OF MODELS WITH PCA AND WITHOUT PCA
# ANAYSIS OF DIFFERENT MODELS (EX RANDOM FOREST HAS VAST DIFFERENCE IN ACCURACY)
```

```{r}
# CHALLENGES

# COULD NOT IMPLEMENT K-MEANS
# CERTAIN ALGOS (SVM) have Taken long time to run
# COORDINATION AND TIME MANAGEMENT DIFFICULTIES (BUT MANAGED)
```