Features first or parameters first

noah977

I have a model that I've developed.

it is based on training an svm on about 30 features.

I understand how to use feature selection with XV to find the best features.
I also understand how to use parameter grid to find the best settings for the SVM.

My question is which to do first.  Is it better to find features first or find the parameters for the SVM first?

Thanks,

land

Hi,
neither nor. You need to optimize both at once since the optimal features are dependend on the SVM's parameter and the other way around.
But probably you might save the efforts with the feature selection, since SVMs cope well with unimportant features. Instead you could try to generate new features...

Greetings,
  Sebastian