MATLAB: Is there a ‘DeltaPredictor’ (output in linear discriminant analysis) equivalent in other classification algorithms (SVM and KNN)

Question

So, with the fitcdiscr classification algorithm, you get an output called 'DeltaPredictor', which indicates predictor importance. Is there an equivalent in the other classification algorithms? Specifically, fitcsvm and fitcknn. If an alternative doesn't exist, how do you calculate the 'DeltaPredictor'? I have tried using other feature selection algorithms (fsrftest and fscmrmr), however, they don't seem to agree with the DeltaPredictor.

Any help will be highly appreciated.

I am not sure if I have given enough information, so if you feel like I haven't, please let me know.

Thank you

Answer 1

Sometimes due to the nature of the underlying model itself, it might not be possible to determine feature importance. For those models where it is possible, this is an known issue, and might be implemented in any future release.

Nevertheless, you can still analyze the feature importance for your classification problem (not specific to SVM) by doing some dimensional reduction or feature extraction.