4
$\begingroup$

The dataSet contains 213 examples of 7 classes . Each example are 25000 features. I want to learn model with SVM (test scenario used are 10-fold cross validation). I am a beginner in machine learning, i want to know what type of kernel i should use in the case of my data: linear or no-linear (as RBF kernel). Also, I want some tips to choose the value of parameter C. Thanks!

$\endgroup$
  • $\begingroup$ Cross-validate. That said, 7 classes from 213 elements is a bit of a tall order... $\endgroup$ – usεr11852 Aug 6 '18 at 23:03
4
$\begingroup$

By elements you mean samples/examples right?

If your dataset consists of 213 samples of 25000 features each you have a much more serious problem to solve than the selection of your SVM hyper-parameters.

In short 213 samples provide are very few for the dimensionality of your problem. In high dimensional spaces you typically need lots of data have a sufficient representation of the space (and this is just one of the problems associated with high-dimensional data). The main issue you have to deal with is feature selection, in order to reduce the number of features.

Now to answer your question:
If I were to choose, I'd take a linear kernel in such a problem. The last thing you need is to project the data to an even higher dimensional space.

$\endgroup$

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.