0
$\begingroup$

I have a dataset composed of some target and non target. I have used Support Vector Machine to classify. I am taking 50% data for training and testing all data.

If I use linear kernel, I get very good classification result. But if I use RBF kernel my resuls are not good. The question is, can we get good result for linear kernel and bad result for RBF kernel for the same dataset?

I was under impression that RBF kernel will also work for linearly seperably data? Can anyone please clarify? I am not ging the details of my work as I assumed it is not relevant to the question, and asking a general point of view?

$\endgroup$
1
  • 1
    $\begingroup$ Are you saying you're using 50% to train and 100% to test? If that is the case, your test performance is not an independent measure. If you mean something else, could you edit your question to clarify? $\endgroup$ – Frans Rodenburg May 14 '19 at 0:02
2
$\begingroup$

Yes, if the data is roughly linearly separable an rbf kernel might perform worse due to over-fitting, especially if the data is unbalanced. The rbf kernel is also more complex and computationally expensive so depending on fitting algorithm, e.g. smo or sgd, and stopping criterion, as well as the amount of hyperparameter optimization, the rbf kernel might fail to reach actual convergence or converge to a sub-optimal solution. If you are using any sort of modern software package, I would guess that over-fitting is a much more probable reason for bad performance.

$\endgroup$
1
  • $\begingroup$ Can we use appropriate parameter in RBK kernel so that the kernal is "close to linear"? $\endgroup$ – Creator May 14 '19 at 1:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.