I am doing a machine learning assignment and I have been asked a question that I have no idea about. I have read my notes and also some posts online, but I can not find anything related. Any idea about how to deal with this?

Having a dataset with labels, where we apply PCA to 2D and 3D as feature transformation, will it improve, reduce or have no effect on performance if we use Logistic Regression to classify the data?

  • $\begingroup$ What do you do once you diagonalize the covariance matrix? $\endgroup$
    – Dave
    Commented Mar 3, 2020 at 22:01
  • $\begingroup$ I multiply the matrix of data and the matrix V (taking only the first n columns, where n is number of features I want) $\endgroup$ Commented Mar 3, 2020 at 22:14
  • $\begingroup$ The answer depends on what you mean by "apply PCA:" could you be specific about what you would do? In particular, how many principal components would you retain? $\endgroup$
    – whuber
    Commented Mar 3, 2020 at 22:25
  • $\begingroup$ Also, is this about performance on in-sample or out-of-sample data? $\endgroup$
    – Dave
    Commented Mar 3, 2020 at 22:26
  • $\begingroup$ See stats.stackexchange.com/questions/596036/… $\endgroup$ Commented Nov 19, 2022 at 2:29


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.