I have a data set where I am doing a binary classification. I have close to 500 features and 200K observations. Now I also have few continuous variables as features.

I don't think just using these features just like that is right. Maybe some sort of transformation (log, sqrt etc.) is needed. How do I find out which transformation is the right transformation. Whether it need to be log, sqrt, squared or kernel function etc.

I do remember there is a technique called Power Transformation. It uses a formula to get the right transformation. Can someone guide here how to find the right transformation of the feature?

  • $\begingroup$ Many threads here on transformation (nearly 900 tagged as such). Did you search before posting? On this information, the main comment is just that your results may be sensitive to whether your variables are transformed. $\endgroup$ – Nick Cox Dec 2 '15 at 0:31
  • $\begingroup$ Sounds like you're talking about a Box-Cox transformation. It's also a pretty common practice to center and scale continuous predictors/features. $\endgroup$ – C.R. Peterson Dec 2 '15 at 1:21
  • $\begingroup$ Yes Box-cox transformation. Exactly I forgot the name. I know there is a formula for that. Any idea how to implement it $\endgroup$ – Baktaawar Dec 2 '15 at 1:33

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.