I have several random forest models that work well. Now, I would like to do feature selection based on this models. How do I find out which features were used frequently by a model e.g. RandomForestClassifier in scikit-learn?
Feature importance it is a property of the random forrest classifier.
See an example here
Try looking at the DALEX package for R. It is designed to specifically investigate the importance of input variables to perhaps otherwise opaque ML algos.