Data Science Stack Exchange is a question and answer site for Data science professionals, Machine Learning specialists, and those interested in learning more about the field. It only takes a minute to sign up.
Sign up to join this community
[Beginner here] If dataset contains - both ordinal, nonordinal (few categories) & nonordinal (multiple categories > 30). Is one supposed to pick one to encapsulate of all such situations or preprocess each type with different encoders?
End goal is to train a ML model.
I would suggest to use different encoders for different types of columns. Lets suppose you have 10 categorical features where 5 of them are ordinal and 5 are nominal. If you one hot encode all of them, you will destroy the sense of order that was in the ordinal features and hence this will affect the performance of the model. Also your dimensionality would increase marginally.
Instead you should use one hot encoder on the nominal features and ordinal encoding techniques on your ordinal features. This will have 2 advantages:-
1.) The sense of order in your data will remain preserved.
2.) Dimensionality will also be less compared to the case where you one hot encode all of them (although that depends on your data).
I would suggest divide your data into different data frames based on the type of features it has and then perform OHE on the nominal feature data frame, Ordinal Encoding on the ordinal feature data frame and then finally concatenate them.
