I have a timeseries dataset with large number of features. Because of that, I need to do PCA. Some of the features are noisy, others are not.

My question is - does it make to do some kind of smoothing (filtering) such as moving average, before doing PCA?

It seems that this might additionally reduce number of eigenvectors needed to represent the original dataset, as the variance will be smaller after smoothing.

  • $\begingroup$ Some discussion here: stats.stackexchange.com/questions/23566 $\endgroup$ – amoeba says Reinstate Monica Dec 14 '16 at 14:59
  • $\begingroup$ Wow, thanks for the super-useful, and very quick link. I was actually expecting that this question will rust here. So, basically, my feeling is good? It does make sense to do the smoothing (such as rolling avg) before PCA? $\endgroup$ – Marko Dec 14 '16 at 15:36

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.