SVM on Spam data

##Uncomment if use this package for the first time

##install.packages("e1071");

library("e1071");

##The Spam data

x<-read.table("spam.Data",sep=",");

n<-nrow(x);

p<-ncol(x);

x[,p]<-as.factor(x[,p]);

colnames(x)[p]<-"class";

##Split the data according to HTF for comparison

T<-3065;

idx2<-sample(1:n,T, replace=FALSE);

spam_train<-x[idx2,];

spam_test<-x[-idx2,];

spam_tune<-tune.svm(class ~., data=spam_train[1:500,],gamma=10^(-6:-2),cost=10^(0:2))

summary(spam_tune);

myGamma<-spam_tune$best.parameters[[1]];

myC<-spam_tune$best.parameters[[2]];

myspam<-svm(class ~ ., data=spam_train,cost=myC, gamma=myGamma,

cross=10, method = "C-classification");

summary(myspam);

myspam_pred<-predict(myspam,spam_test,decision.values=TRUE);

conf<-table(myspam_pred,spam_test$class);

acc<-sum(diag(conf))/sum(conf);

cat("The accuracy on the test set is", acc,"\n");