Feature Importance value for SVM, KNN, and Discriminant type classifier?

I am using MATLAB classification learner app for my classification task. I am interested to know feature importance value for each kind of machine learning model. I know feature importance for tree and ensemble type models can be find out by 'predictorimportance' function. Is there any way that can be used to determine feature importance value in case of SVM,KNN Discriminant type classifier?

Answer 1

[Ayush Aniket]

In MATLAB, there is no direct method to find out feature importance for models like SVM, KNN, and discriminant analysis.

However, for determining feature importance for these models you can use Recursive Feature Elimination (RFE) technique that recursively removes the least important features based on the model's performance and can be adapted to any of the above-mentioned models.

You can use the sequentialfs function in MATLAB to implement it. The function uses a custom criterion function to select the features and these can be the models SVM, KNN etc. Refer to the following documentation to read about the process of using a custom criterion function:

https://www.mathworks.com/help/stats/sequentialfs.html#description

The importance of the feature can be implied from the order in they are removed.