We know that svm considers a weight for every dimension of input (vector w), so why we use feature selection or extraction and reduce the dimensions? Isn't it right that if a feature is not important, its w in svm is so low or near to zero?

  • $\begingroup$ They do not help. $\endgroup$ – Frank Harrell Jul 13 '17 at 14:06
  • $\begingroup$ @FrankHarrell hmm, what's the point of using them? $\endgroup$ – user137927 Jul 13 '17 at 14:14
  • $\begingroup$ Parsimony is the enemy of predictive discrimination. Parsimony is used for its own sake to make things (falsely) appear simpler. It is not used to improve predictions. $\endgroup$ – Frank Harrell Jul 13 '17 at 16:10

Your Answer

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

Browse other questions tagged or ask your own question.