Lab of Support Vector Machines with Kernel Methods

Goal

The performance of the linear SVM obtained in the previous lab is decent, though not the best, so there could be some room for improvement here. That’s what we want to investigate in this lab: if using other kernels may help boost the performance of SVM.

We’ll continue to use the kernlab and caret packages in this lab.

Brief refresher

One approach to make classification easier is to sparsify the data by elevating them to a higher-dimensional space, with the hope of making classes more separable. The kernel trick is a technique that does that efficiently by bypassing explicit project steps.

SVM, among other techniques, can be enhanced with the kernel trick to compute nonlinear decision boundaries.

Preparation