Quadratic SVM for feature selection

Dear Community,

I have 9 observations, 23 features and two classes (low grade and high grade). I am trying to find a combination of these features that can help with the classification.

I am quite new at these classification problems so I decided to run the Classification Learner App. I decided to run in parallel all the models and the quadratic SVM showed the highest accuracy. Then I tried to run it outside the app with the two variables: TABLE (with the features) and grade (with the labels).

I don't really understand the results as I was expecting a 'score' for each feature so as to reduce the number of features for the classification.

Is the hypothesis wrong? Shall I use something else (as the LASSO for example)?

Thank you for the help.

Answer 1

[Aditya Patil]

In general, SVM can't be used to determine feature importance. You can read more about this in this answer. There are various feature selection and extraction techniques available in MATLAB, which you can read about in the dimensionality reduction doc page.