fearture_importances_

[bixiang]

Hello, I have one question about kaggle competition. Some one told me a trick: through using fearture_importances_ of tree model , we can remove the features whose importance are weak or add compound features using the important, then ,we can train model again. All of these can get better performance. I tried these tricks, but the performance was not better. Could you tell me Why this happened? Thank you very much!!!