SVM_Hyperparametertuning.Rmd

---
title: "Support Vector Machines"
author: "Sushama_Rangarajan"
output:
  pdf_document: default
  html_document: default
---

```{r setup, include=FALSE}
if(!require("pacman")) install.packages("pacman")
pacman::p_load(e1071, ggplot2, caret, rmarkdown, corrplot)
theme_set(theme_classic())
```

Reading the data and partitioning it with 80% train and 20% test

```{r dataloading}
j1 <- read.csv("juice.csv")
j1$Purchase <- as.factor(j1$Purchase)
```

```{r Partitioning}
set.seed(123)
part <- createDataPartition(j1$Purchase, p=0.8, list = FALSE)
j1.train <- j1[part,]
j1.test <- j1[-part,]
```

# Linear Kernel with cost = 0.01

```{r Linear SVM}
svm1 <- svm(Purchase~., data=j1.train,kernel = "linear",cost=0.01,type = "C-classification")
summary(svm1)
```


Predicting the test and train dataset with linear SVM 

```{r Predicting with linear SVM}
#prediction of test data

pr1 <- predict(svm1,j1.test)
conf.test <- table(Predicted = pr1, Actual = j1.test$Purchase)
conf.test

sensitivity(conf.test)
specificity(conf.test)
#Test Error rate
(1- (sum(diag(conf.test)) / sum(conf.test)))

#prediction of train data

pr11 <- predict(svm1,j1.train)
conf.train<- table(Predicted = pr11, Actual = j1.train$Purchase)
conf.train

#Train Error rate
(1- (sum(diag(conf.train)) / sum(conf.train)))
```



Hyperparameter Tuning:

```{r Tuning Linear model}

set.seed(123)
#sequence was modified after multiple iterations
tune.linear <- tune(svm, Purchase ~.,data = j1.train, kernel = "linear", ranges = list(cost = seq(0.01,10.1,by=0.3)))
summary(tune.linear)
plot(tune.linear)
best1 <- tune.linear$best.model
summary(best1)
```

#Prediction:
```{r Predicting with best linear model}
#Prediction using test data
bestpred1 <- predict(best1,j1.test)
conf.test1 <- table(predicted = bestpred1, actual = j1.test$Purchase)
conf.test1

#Test Error rate
(1- (sum(diag(conf.test1)) / sum(conf.test1)))

#Prediction using train data
bestpred2 <- predict(best1,j1.train)
conf.train1 <- table(predicted = bestpred2, actual = j1.train$Purchase)
conf.train1

#Train Error rate
(1- (sum(diag(conf.train1)) / sum(conf.train1)))

```



```{r Radial SVM}
set.seed(123)
svm2<- svm(Purchase~., data=j1.train, cost = 0.01)
summary(svm2)
```



Prediction : 

```{r Predicting with Radial SVM}
#prediction using test data
pr2 <- predict(svm2,j1.test)
conf.test2 <- table(Predicted = pr2, Actual = j1.test$Purchase)
conf.test2

#Test Error rate
(1- (sum(diag(conf.test2)) / sum(conf.test2)))

#prediction using train data
pr21 <- predict(svm2,j1.train)

conf.train2<- table(Predicted = pr21, Actual = j1.train$Purchase)
conf.train2

#Train Error rate
(1- (sum(diag(conf.train2)) / sum(conf.train2)))

```

Hyperparameter Tuning

```{r Radial - Tuning}

set.seed(123)
#sequence was modified after multiple iterations
tune.radial <- tune(svm, Purchase ~.,data = j1.train, kernel = "radial", ranges = list(cost = seq(0.01,10.1,by=1)))
summary(tune.radial)
plot(tune.radial)
best2 <- tune.radial$best.model
summary(best2)

```


Prediction
```{r Predicting with Radial best model}
#Prediction using test 
bestpred2 <- predict(best2,j1.test)
conf.test3 <- table(predicted = bestpred2, actual = j1.test$Purchase)
conf.test3

#Test Error rate
(1- (sum(diag(conf.test3)) / sum(conf.test3)))

#Prediction using train 
bestpred3 <- predict(best2,j1.train)
conf.train3 <- table(predicted = bestpred3, actual = j1.train$Purchase)
conf.train3

#Train Error rate
(1- (sum(diag(conf.train3)) / sum(conf.train3)))

```



# Polynomial Kernel with degree =2 and cost = 0.01
```{r Polynomial SVM}
svm3<- svm(Purchase~., data=j1.train,kernel = "polynomial",degree = 2, cost = 0.01)
summary(svm3)

```


```{r Predicting with Polynomial SVM}
#Prediction using test 
pr3 <- predict(svm3,j1.test)
conf.test4 <- table(Predicted = pr3, Actual = j1.test$Purchase)
conf.test4

#Test Error rate
(1- (sum(diag(conf.test4)) / sum(conf.test4)))

#prediction of train data
pr31 <- predict(svm3,j1.train)

conf.train4<- table(Predicted = pr31, Actual = j1.train$Purchase)
conf.train4

#Train Error rate
(1- (sum(diag(conf.train4)) / sum(conf.train4)))
```


Hyperparameter Tuning

```{r Polynomial - Tuning}
#Hyper parameter Tuning 
set.seed(123)
#sequence was modified after multiple iterations
tune.poly <- tune(svm, Purchase ~.,data = j1.train, kernel = "polynomial", degree = 2, ranges = list(cost = seq(0.01,10.1,by=0.5)))
summary(tune.poly)
plot(tune.poly)
best3 <- tune.poly$best.model
summary(best3)
```


```{r Predicting with best polynomial model}
#Prediction using test 
bestpred4 <- predict(best3,j1.test)
conf.test5 <- table(predicted = bestpred4, actual = j1.test$Purchase)
conf.test5

#Test Error rate
(1- (sum(diag(conf.test5)) / sum(conf.test5)))

#Prediction using train 
bestpred5 <- predict(best3,j1.train)
conf.train5 <- table(predicted = bestpred5, actual = j1.train$Purchase)
conf.train5

#Train Error rate
(1- (sum(diag(conf.train5)) / sum(conf.train5)))

```