Discriminant factors in SVM procedure


Dear friends,
I'm looking for a strategy to detect the most discriminant predictors in a SVM model, as for example recursive feature elimination or others. Is it possible in Knime?



I've done this before using Weka, and as Weka is integrated in KNIME, it should be possible as well. In WEKA, there's something called the SVMAttributeEval, which you can load in a KNIME node which wraps a WEKA metaclassifier, such as the AttributeSelectedClassifier.

If you use the Ranker search method in the above node, you'll get a list of your most significant discriminant predictors.




You could also use the Backward Feature Elimination meta-node.

I've found the meta-node...




I'll explore the two strategies that you propose.

In the case of the meta-node of BFE, where is ubicated?

Thanks a lot.