For a classification problem with numerical predictors and a categorical multi-class response I am trying to use Recursive Feature Elimination with Random Forests to identfy relevant features out of a data set.
I am using scikit-learn for this purpose, which usually required target variables to be either integer for ordinal categories or one hot encoded for non-ordinal categorical classes. The
fit() method for both
RFECV in scikit-learn however accepts only integer target variables.
Encoding my classes with integers is not a problem but I am unsure if this is correct. Can I use a integer encoded target for a non-ordinal classification problem? Or could this cause problems because the model is using the order of classes that is artificially created?