I have a random forest classifier which gave me a feature importance rank.
How can I derive statistical significance of the important features, similar to a regression model where you can infer statistical significance of the betas?
Asked
Active
Viewed 523 times
1
-
Do you want to use a parametric or non-parametric test ? e.g. t-test or permutation testing? – seralouk Jul 24 '18 at 08:19
1 Answers
0
Your question is a bit too broad and unclear.
An easy way you can look at the feature_importance_values as percentage is by normalizing their values:
importance_sum = sum(clf. feature_importances_)
feature_importance_as_percent = [100*(x/sum) for x in clf.feature_importances_]
Other methods would involve parametric or non-parametric tests.
Read also this: How are feature_importances in RandomForestClassifier determined?
seralouk
- 30,938
- 9
- 118
- 133