5
$\begingroup$

Case:
I have a dataset with a large number of features, which I want to reduce. Should I look for a method that identifies the most important ones and throw away the rest, or should I look for a method that attempts to create a smaller number of new features from the old ones?

Which method of dimensionality reduction is more common in practice: feature selection or feature extraction? Is one superior to the other? How should I know which one to prefer? Finally, is there any benefit in combining the two?

$\endgroup$
  • $\begingroup$ Have you tried PCA? $\endgroup$ – Bayequentist Sep 16 '18 at 23:51
  • $\begingroup$ Welcome to CV. I don't think this question can be answered without some more background information. Both approaches are common, but are used to different ends. For example, does it matter to you how well you can interpret the effects of individual input variables at the end? $\endgroup$ – Frans Rodenburg Sep 17 '18 at 0:34
2
$\begingroup$

It depends on the situation and objective, however in general it is preferable to retain your original variables as these can be more easily interpreted. There is no better option between the two as it depends entirely on the problem at hand.

Combining the two is unnecessary and you should pick one. And this is about as much that can be said given the details that you've given us.

$\endgroup$
  • $\begingroup$ I was wondering in general if one is preferred to the other. Your answer covered exactly that. Thanks $\endgroup$ – ILM91 Sep 19 '18 at 22:01

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.