How to choose the right kernel features

Question

I have a very general question: how to choose the right kernel function for SVM? I know that the final answer is to try all the kernels, perform an off-sample check and select one of them with the best classification result. But other than that, are there any recommendations for using various kernel functions?

Answer 1

Take a look to find the answer.

https://stats.stackexchange.com/questions/18030/how-to-select-kernel-for-svm

In principle, there is not one good way to choose if you do not know something important about your data that could determine the correct kernel to use. However, follow the link above for more specific information.

Answer 2

Try a linear core first, simply because it is much faster and can give great results in many cases (in particular, problems with large sizes).

If the linear core fails, overall your best bet is the RBF core. They are known to cope with a lot of problems very well.