I have data that I will use as a feature to Elastic Net. I thought I should transform the data using either Box Cox or Yeo Johnson.
The transformed data looks weird and I'm not sure if I should transform the data or not.
enter image description here

Besides the large bar at -2 for Yeo Johnson the distribution looks ok, but I'm not sure if large amount of values around -2 will affect model performance or not.

So my question is, is it always safe to transform the data? I.e. can it affect model performance, in a negative way, by transforming the data?

  • $\begingroup$ Some implementations normalize the data by default, so be careful not to overnormalize. Standardization is often the #1 choice. $\endgroup$ – user2974951 Nov 7 '19 at 9:56
  • $\begingroup$ You do not need normal distributions for predictor variables. (You do not need a normal distribution for the response variable, either. When regressions make a normality assumption, it is about the error term, not any of the marginal distributions.) $\endgroup$ – Dave Oct 19 '20 at 1:59

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.