Modelling 1.Rmd

---
title: "KPMG Thesis Project"
output: html_document
---
#Work on Original Dataset

## Data Reading:
```{r}
library(readr)

#uncomment when you want to read the data from fresh
Data <- read_delim("Data/KPMGDT.csv",",", escape_double = FALSE, trim_ws = TRUE)
```

## Data Stats
```{r}
dim(Data)
names(Data)
summary(Data)
```


##Subsetting Train and Test Datasets
```{r}
set.seed(100)
dataPartition = sample(2,nrow(Data),replace=TRUE,prob=c(0.7,0.3))
trainData <- Data[dataPartition ==1,]
testData <- Data[dataPartition ==2,]
```


##Data Prepration
```{r}
Data$CreditRating <- as.factor(Data$CreditRating)
Data$InterestType <- as.factor(Data$InterestType)
Data$MortgageType <- as.factor(Data$MortgageType)
Data$NewLoan <- as.factor(Data$NewLoan)
Data$ProbationaryLoans <- as.factor(Data$ProbationaryLoans)
Data$LTVCategory <- as.factor(Data$LTVCategory)
Data$InArrears <- as.factor(Data$InArrears)
Data$County <- as.factor(Data$County)
Data$DefaultedLoans <- as.factor(Data$DefaultedLoans)
str(Data)
```
##GLM

###Building Simple GLM Model
```{r}

simpleglm <- glm(DefaultedLoans ~ CreditRating + InterestIncome + 
    log(PropertyValue) + log(LoanBalance) + AnnualPYMT + LTV + 
    InterestType + NewLoan + ProbationaryLoans + MortgageYears + 
    MortgageType + InArrears + County + AddressLatitude + AddressLongitude, 
    family = "binomial", data = trainData)

summary(simpleglm)


```
###Running simple GLM Model on test data of predict loan default

```{r}
testData$prediction = predict(simpleglm, newdata=testData, type="response")
```

###Testing Accuracy of glm() model
```{r}
table(testData$DefaultedLoans, as.numeric(testData$prediction >= 0.5))
```
### Accuracy
```{r}
(67566+3154)/nrow(testData)
```
### ROC Curve
```{r}
library(ROCR)
#roc_prediction = prediction(testData$prediction, testData$DefaultedLoans)
#as.numeric(performance(pred, "auc")@y.values)

# Make predictions on training set
train_pred = predict(simpleglm, type="response")

# Prediction function
ROCT_Prediction = prediction(train_pred, trainData$DefaultedLoans)

# Performance function
ROCR_Performance = performance(ROCT_Prediction, "tpr", "fpr")

# Plot ROC curve
plot(ROCR_Performance)

# Add colors
plot(ROCR_Performance, colorize=TRUE)

# Add threshold labels 
plot(ROCR_Performance, colorize=TRUE, print.cutoffs.at=seq(0,1.5,by=0.1), text.adj=c(-0.2,1.7))
```

## Vizualization of Ireland property price market

```{r}
## Load the library
library(dplyr)
library(maps)
library(reshape2)
library(leaflet)
library(ggplot2)
library(ggmap)
library(gridExtra)
library(htmlwidgets)
library(readr)
weatherIcon <- makeIcon(
                         iconUrl = "./fig/weather.png",
                         iconWidth = 30, 
                         iconHeight = 30
                        )

popupInfo <- paste(testData[['ContractRef']], 
                   ", ", 
                   testData[['LoanBalance']],
                   "<br>", 
                   "Average January Temp in F:  ",
                   testData[['DefaultedLoans']],
                   "<br>", 
                   "Credit Rating:  ",
                   testData[['CreditRating']],
                   "<br>", 
                   "LTV:  ",
                   testData[['LTV']],
                   "<br>", 
                   "Property Value:  ",
                   testData[['PropertyValue']],
                   sep='')


MapDisplay <- leaflet(testData) %>% 
                          setView(-6.24420, 53.30867, zoom = 12) %>%
                                 addTiles() %>%
                                      addMarkers(testData$AddressLongitude, testData$AddressLatitude, popup= ~ popupInfo,
                                      options = popupOptions(closeButton = TRUE),
                                      clusterOptions = markerClusterOptions(), 
                                      icon = weatherIcon)
#MapDisplay
saveWidget(MapDisplay, file="E:/KPMG/KPMG/Output/MapDisplay.html")

```
####################################
#
#Update this part
#
#####################################

```{r}
library(lubridate)
library(ggplot2)
library(dplyr)
library(stringr)
library(caret)
library(rpart)
library(rattle)
library(ROSE)
library(ROCR)
library(MASS)
library(ipred)
library(plyr)
library(rpart.plot)
library(readr)
levels.default(testData$DefaultedLoans)
table(testData$DefaultedLoans, testData$CreditRating)
ggplot(testData, aes(x = testData$LoanBalance)) +geom_histogram(aes(fill = CreditRating)) +facet_wrap(~DefaultedLoans, ncol=1)

```
```{r}
index = createDataPartition(y = testData$DefaultedLoans, p = 0.90)[[1]]
loans.sample <- testData[-index,]
ggplot(loans.sample, aes(x = testData$CreditRating, y = testData$CreditRatingMovement)) + geom_point(aes(color = testData$MortgageYears))

```



```{r}
predictions.1 <- (predict(mydata.rpart.1, mydata.test, type = "class"))
confusionMatrix(predictions.1, mydata.test$DefaultedLoans)
```


#Model for Tableau
```{r}
#mydata <- read.csv("E:/KPMG/Data/Prep/Sample1_3.csv")
lrmodel <- glm(DefaultedLoans ~ CreditRating + LoanBalance + PropertyValue, data = trainData, family = "binomial");
save(lrmodel, file = "Model/mymodel.rda")

```

```{r}
load("mymodel.rda")
prob <- predict(lrmodel, newdata = testData, type = "response")
plot(prob,testData$CreditRating)

```

#Decision Tree for Tableu
```{r}
#Decision Tree for Tableu
library(rpart)
library(rattle)					# Fancy tree plot
library(rpart.plot)				# Enhanced tree plots
library(RColorBrewer)				# Color selection for fancy tree plot
library(party)					# Alternative decision tree algorithm
library(partykit)				# Convert rpart object to BinaryTree
library(caret)
fit <- rpart(DefaultedLoans ~ NewLoan+County+LoanBalance + PropertyValue+ CreditRating+ AnnualPYMT,method = "class",data=testData, control =  rpart.control(minisplit=5,cp = 0.001))

save(fit, file = "Output/myclassificationtree.rda")
print(fit)
prp(fit)
tree.1 <- fit
fancyRpartPlot(tree.1)
```

## Confusion Matrix for traindata
```{r}
defaultPredict <- predict(fit, trainData, type = 'class')
matrix <- table(defaultPredict, trainData$DefaultedLoans)
print(matrix)
print((matrix[4]+matrix[1])/nrow(trainData))
```

## Classification Tree Model for Tableau
```{r}
library(rpart)
library(rpart.plot)				# Enhanced tree plots
library(RColorBrewer)
library(rattle)					# Fancy tree plot)				# Color selection for fancy tree plot
library(party)					# Alternative decision tree algorithm
library(partykit)				# Convert rpart object to BinaryTree
library(caret)
fit <- rpart(DefaultedLoans ~ CreditRating+ County+ LoanBalance +PropertyValue+ NewLoan + ProbationaryLoans+LTV + LTVCategory +InterestIncome+ AddressLatitude, data=testData, control=rpart.control(minsplit =5,cp=0.001),method="class")

prp(fit)
```