Modelling 2.Rmd

---
title: "Analysis v2"
author: "Deepak & Shruti"
date: "30 July 2017"
output:
  html_document: default
  pdf_document: default
---

#Work on Original Dataset

## Data Reading:
```{r}
library(readr)
Datav2 <- read_delim("Data/finalData.csv", 
    ";", escape_double = FALSE, trim_ws = TRUE)
```
## Data Stats
```{r}
Datav2$InterestType <- as.factor(Datav2$InterestType)
Datav2$MortgageType <- as.factor(Datav2$MortgageType)
Datav2$NewLoan <- as.factor(Datav2$NewLoan)
Datav2$ProbationaryLoans <- as.factor(Datav2$ProbationaryLoans)
Datav2$LTVCategory <- as.factor(Datav2$LTVCategory)
Datav2$InArrears <- as.factor(Datav2$InArrears)
#Datav2$County <- as.factor(Datav2$County)
Datav2$DefaultedLoans <- as.factor(Datav2$DefaultedLoans)
m1 <- Datav2
m2 <- Datav2
m3 <- Datav2

```


##Data Prepration
```{r}
#Datav2$CreditRating <- as.factor(Datav2$CreditRating)
Datav2$InterestType <- as.factor(Datav2$InterestType)
Datav2$MortgageType <- as.factor(Datav2$MortgageType)
Datav2$NewLoan <- as.factor(Datav2$NewLoan)
Datav2$ProbationaryLoans <- as.factor(Datav2$ProbationaryLoans)
Datav2$LTVCategory <- as.factor(Datav2$LTVCategory)
Datav2$InArrears <- as.factor(Datav2$InArrears)
#Datav2$County <- as.factor(Datav2$County)
Datav2$DefaultedLoans <- as.factor(Datav2$DefaultedLoans)
#Datav2$LoanBalance <- scale(Datav2$LoanBalance)
#Datav2$PropertyValue <- scale(Datav2$PropertyValue)
#Datav2$InterestIncome <-scale(Datav2$InterestIncome)
#Datav2$AnnualPYMT <-scale(Datav2$AnnualPYMT)
#str(Datav2)
```





##Subsetting Train and Test Datasets
```{r}
set.seed(100)
dataPartition = sample(2,nrow(Datav2),replace=TRUE,prob=c(0.7,0.3))
trainDatav2 <- Datav2[dataPartition ==1,]
testDatav2 <- Datav2[dataPartition ==2,]
```

##GLM

### Model trained with all variables
```{r}

#m2 <- glm(DefaultedLoans ~., family = "binomial", data = trainDatav2)
#step(m2)

```

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

simpleglmv2 <- glm(DefaultedLoans ~ CreditRating + InterestIncome + 
    PropertyValue + LoanBalance + AnnualPYMT + LTV + 
    InterestType + NewLoan + ProbationaryLoans + MortgageYears + 
    MortgageType + InArrears + County + AddressLatitude + AddressLongitude, 
    family = "binomial", data = trainDatav2)

summary(simpleglmv2)

save(simpleglmv2, file = "Model/simpleglmv2.rda")

```


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

```{r}
testDatav2$prediction <- predict(simpleglmv2, newdata=testDatav2, type="response")
```

###Testing Accuracy of glm() model
```{r}
table(testDatav2$DefaultedLoans, as.numeric(testDatav2$prediction >= 0.5))

```
### Accuracy
```{r}
(868+8190)/nrow(testDatav2)
(7849+2953)/nrow(testDatav2)
```
### ROC Curve
```{r}
library(ROCR)
#roc_prediction = prediction(testDatav2$prediction, testDatav2$DefaultedLoans)as.numeric(performance(pred, "auc")@y.values)

# Make predictions on training set
train_pred = predict(simpleglmv2, type="response")

# Prediction function
ROCT_Prediction = prediction(train_pred, trainDatav2$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,by=0.05), text.adj=c(-0.2,2.0))
```


## 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(testDatav2[['ContractRef']], 
                   ", ", 
                   testDatav2[['LoanBalance']],
                   "<br>", 
                   "Has Defaulted?:  ",
                   testDatav2[['DefaultedLoans']],
                   "<br>", 
                   "Credit Rating is::  ",
                   testDatav2[['CreditRating']],
                   "<br>", 
                   "LTV:  ",
                   testDatav2[['LTV']],
                   "<br>", 
                   "Property Value:  ",
                   testDatav2[['PropertyValue']],
                   sep='')


MapDisplay <- leaflet(testDatav2) %>% 
                          setView(-6.24420, 53.30867, zoom = 12) %>%
                                 addTiles() %>%
                                      addMarkers(testDatav2$AddressLongitude, testDatav2$AddressLatitude, popup= ~ popupInfo,
                                      options = popupOptions(closeButton = TRUE),
                                      clusterOptions = markerClusterOptions(), 
                                      icon = weatherIcon)
MapDisplay
saveWidget(MapDisplay, file="MapDisplayv2.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(trainDatav2$DefaultedLoans)
table(trainDatav2$DefaultedLoans, trainDatav2$CreditRating)
ggplot(trainDatav2, aes(x = trainDatav2$LoanBalance)) +geom_histogram(aes(fill = CreditRating)) +facet_wrap(~DefaultedLoans, ncol=1)

```

```{r}
#Decision Tree
mydata.rpart.0 <- rpart(DefaultedLoans ~ ., data = trainDatav2)

mydata.rpart.1 <- rpart(DefaultedLoans ~ . , data = testDatav2, 
                      control=rpart.control(minsplit=10, minbucket = 3, cp=0.0006))


fancyRpartPlot(mydata.rpart.1)
```

```{r}
predictions.1 <- (predict(mydata.rpart.1, testDatav2, type = "class"))
confusionMatrix(predictions.1, testDatav2$DefaultedLoans)
```

#Model for Tableau
```{r}
lrmodel <- glm(DefaultedLoans ~ CreditRating + LoanBalance + PropertyValue, data = trainDatav2, family = "binomial");
save(lrmodel, file = "Model/lrmodelv2.rda")

```

```{r}
load("Model/lrmodelv2.rda")
prob <- predict(lrmodel, newdata = testDatav2, type = "response")
plot(prob,testDatav2$CreditRating)

```
#Decision Tree

##Decision Tree Model
```{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)
library(tree)
defaultLoanTree <- rpart(DefaultedLoans ~ NewLoan + County + LoanBalance + PropertyValue + InterestIncome + CreditRating + AnnualPYMT + County + LTV + LTVCategory + InArrears + MortgageType + MortgageYears + AddressLatitude + AddressLongitude ,method = "class",data=trainDatav2, control =  rpart.control(minisplit=5000,cp = 0.001))

save(fit, file = "Model/classificationTreeV2.rda")
print(defaultLoanTree)
prp(defaultLoanTree)
tree.1 <- defaultLoanTree
fancyRpartPlot(tree.1)

##################################

## Prunning Decision Tree using rpart()

#PruneTree = prune(fit, cp = 0.003)
#prp(PruneTree)
#fancyRpartPlot(PruneTree)
```
## Stats for Decision Tree
```{r}
printcp(defaultLoanTree)
```

## Confusion Matrix for traindata
```{r}
defaultPredict <- predict(defaultLoanTree, trainDatav2, type = 'class')
matrix <- table(defaultPredict, trainDatav2$DefaultedLoans)
print(matrix)
print((matrix[4]+matrix[1])/nrow(trainDatav2))
```
## Confusion Matrix for test data
```{r}
defaultPredict <- predict(defaultLoanTree, testDatav2, type = 'class')
matrix <- table(defaultPredict, testDatav2$DefaultedLoans)
print(matrix)
print((matrix[4]+matrix[1])/nrow(testDatav2))
```


## Classification Tree Model (Small Tree)
```{r}
library(rpart)
library(rpart.plot)				# Enhanced tree plots
library(RColorBrewer)
library(rattle)					# Fancy tree plot)				# Color selection for fancy tree plot
library(caret)
fit <- rpart(DefaultedLoans ~ NewLoan + County + LoanBalance + 
    PropertyValue + InterestIncome + CreditRating + AnnualPYMT + 
    County + LTV + LTVCategory + InArrears + MortgageType + MortgageYears + 
    AddressLatitude + AddressLongitude, data=testDatav2, control=rpart.control(minsplit =25,cp=0.002),method="class")
prp(fit)
fancyRpartPlot(fit)

```

## Decision Tree using party() package
```{r}
library(party)
library(partykit)
fit <- ctree(DefaultedLoans ~ NewLoan + County + LoanBalance + 
    PropertyValue + InterestIncome + CreditRating + AnnualPYMT + 
    County + LTV + LTVCategory + InArrears + MortgageType + MortgageYears + 
    AddressLatitude + AddressLongitude, data=testDatav2)
plot (fit, main="Conditional Inference Tree")  # the ctreeprp(fit)

```
#Adding Prediction variable in Dataset of Decision Tree

```{r}
dt_data_model <-testDatav2
dt_data_model$PredictionDT <- predict(defaultLoanTree, testDatav2, type = 'class')
write.csv(dt_data_model, "Data/dt_data_model.csv")
```