SVM

Question 1

Pre processing required for SVM Classification

Answer 1

you have to make the response variable a factor
y = as.factor(y)

You can scale the variables if needed by using a parameter in the svm function svm(…, scale = T)

Question 2

how to specify the shape of the decision boundary

Answer 2

svm(y~., data=dat, kernel = ‘linear’…)

you can specify different values for the kernel

Question 3

function used to tune svm model

Answer 3

library(e1071)

tune(svm, y~., data = dat, kernel = ‘linear’, ranges = list(cost(c(0.001, .01, .1, 1, 5, 10 , 100)))

Question 4

what package has ROC curves

Answer 4

library(ROCR)

Question 5

What do you want to do when p&raquo_space; n

Answer 5

use a linear decision boundary

Question 6

how to handle a multi-class response variable in svm

Answer 6

svm() function in the e1071 library will take care of it, just have to make sure that response variable is a factor. can set it to factor using y=as.factor(y)