I have been using WEKA to classify very long duration audio recordings. The best performing classifiers have been Random Forest and J48. The attributes used to classify the audio are acoustic indices. This process of generating these indices is quite resource intensive.

I would like to determine the importance of the various attributes. Is there a way for these classifiers to report this?

I see J48 produces a decision tree, is it safe to day the attributes used at the root of the tree are most important?


1 Answer 1


If you must use WEKA, I would suggest looking in the Attribute Selection panel. In this panel, you can rank attributes by information gain, as well as look at which subsets of attributes perform best. To my knowledge, you can't get this information directly from the classifier in the WEKA Explorer, but it is quite easy to get in other frameworks such as scikit-learn: see this example in the docs.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.