I have this feature for which the distribution looks like: enter image description here

I am curious as to what kind of feature processing I can do (and to similar other features) which makes it amenable for kNN.

Here's what I tried:

In an attempt to convert the feature from exponential (here) to normal distribution, I did the ecdf = exponential.CDF(X) and then normal.inverseCDF(ecdf). But the resulting distribution is also exponential in nature.

In terms of feature pre-processing, what else can I do?

  • $\begingroup$ Are you sure that the data is valid? For instance that the 0 values don't mean 'missing value', or has been entered in a different unit (ex millimeters instead of meters). $\endgroup$ – jonnor Oct 26 '18 at 13:36

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.