library(e1071)
library(caret)
## Since SVM only accepts numeric variables, encoding the famhist feature
train.encoded_row <- apply(data.frame(train.data$famhist), 1, function(x) ifelse(x == "Present", 1, 0))
train.data["famhist"] <- train.encoded_row


validation.encoded_row <- apply(data.frame(validation.data$famhist), 1, function(x) ifelse(x == "Present", 1, 0))
validation.data["famhist"] <- validation.encoded_row



svm.model <- svm(chd ~ ., data=train.data, type="C-classification")
svm.pred <- predict(svm.model, validation.data)
confusionMatrix(svm.pred, validation.data$chd)

library(ROCR)
svm.roc <- prediction(predictions=c(svm.pred), labels=c(validation.data$chd))
svm.auc <- performance(svm.roc, 'tpr', 'fpr')
plot(svm.auc, colorize=T, main="Svm ROC")




library(nnet)
library(caret)


## one hot encoding to use softmax
train.Y0 = train.data$chd == 0
train.Y1 = train.data$chd == 0

## Neural network with a single hidden layer
nn1=nnet(train.data, cbind.data.frame(train.Y0, train.Y1), size=20, maxit = 500, softmax = TRUE)

## get the prediction
nn.predictions.encoded = predict(nn1, data=validation.data,type = "class")
nn.pred = apply(as.data.frame(nn.predictions.encoded), 1, function(x) ifelse(x == "train.Y1", 1, 0))
confusionMatrix(nn.pred, validation.data$chd)

library(ROCR)
nn.roc <- prediction(predictions=c(nn.pred), labels=c(validation.data$chd))
nn.auc <- performance(nn.roc, 'tpr', 'fpr')
plot(nn.auc, colorize=T, main="Neural Net ROC")