There are categorical features which have two different value in my dataframe next to numerical features. I've converted these categorical values to 0 or 1.

I will apply correalation elimination on features after calculating correlation coefficients. Depending on type of features, methods are given below:

Numeric - Numeric: Pearson

Numeric - Categoric: Cramer_V

Categoric - Categoric: Correlation Ratio

That's why I could not be sure what should be type of converted categorical features? Numerical or categorical ?

Another reason to I asked this question is that when I create dummy features for the categorical features which have only two different values, it creates features contains 0 and 1 like how I did manually. So after this process it's taking these features as numerical. But still each value from the feature represents a class and I think feature type should not be numerical.

  • $\begingroup$ I've converted these categorical values to 0 or 1. How and what's the procedure adopted by you ? $\endgroup$ Commented Apr 5, 2020 at 4:14
  • $\begingroup$ Should the type of Boolean categorical features be numerical or categorical after encoding? What do you mean by Boolean categorical features ? Translating categories into Boolean- is it valid ? $\endgroup$ Commented Apr 5, 2020 at 4:43
  • $\begingroup$ I meant to convert categorical values with two classes into numerical ones. For example, when we convert the "Gender" variable to numerical, we get 0 and 1, so there is no problem here. What should be the correlation approach for the gender variable converted into numerical in the next step? Should it be categorically treated in correlation or should it be treated numerically because it is converted into a numerical type? $\endgroup$
    – talatccan
    Commented Apr 7, 2020 at 8:44


Your Answer

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