SVM

rm(list=ls(all=TRUE))

#install.packages("mlbench")
require(mlbench)
require(e1071)

#uploading data
data("BreastCancer")
BC <- subset(BreastCancer, select=-Id) 
table(is.na(BC))
BC<-na.omit(BC)

#diving into training and test set
set.seed(1)
train=sample(nrow(BC),nrow(BC)*0.7)

model <- svm(Class ~ ., data=BC[train,], cross=10,scale=TRUE) 
summary(model)
BC<-as.data.frame(BC,na.omit=TRUE)
ypred=predict(model,BC[-train,])
sum(diag(table(predict=ypred, truth=BC[-train,10])))/nrow(BC[-train,])

#repeating the procedure 50 times
accuracy<-c()
for(i in 1:50){
  train=sample(nrow(BC),nrow(BC)*0.7)
  model <- svm(Class ~ ., data=BC[train,], cross=10,scale=TRUE) 
  BC<-as.data.frame(BC,na.omit=TRUE)
  ypred=predict(model,BC[-train,])
  accuracy[i]=sum(diag(table(predict=ypred, truth=BC[-train,10])))/nrow(BC[-train,])
}
mean(accuracy)
sd(accuracy)

mean_t<-c()
sd_t<-c()
for(j in seq(from=0.5,to=0.95,by=0.05)){
  accuracy<-c()
  for(i in 1:50){
    train=sample(nrow(BC),nrow(BC)*j)
    model <- svm(Class ~ ., data=BC[train,], cross=10,scale=TRUE) 
    BC<-as.data.frame(BC,na.omit=TRUE)
    ypred=predict(model,BC[-train,])
    accuracy[i]=sum(diag(table(predict=ypred, truth=BC[-train,10])))/nrow(BC[-train,])
  }
  mean_t[length(mean_t)+1]=mean(accuracy)
  sd_t[length(sd_t)+1]=sd(accuracy)
}

0.35*50000
50000*18+1

require(ggplot2)
ggplot()+geom_point(aes(x=seq(from=0.5,to=0.95,by=0.05),y=mean_t))+ggtitle("Mean accuracy v. Data split")