Unbiased Measurement of Feature Importance in Tree-Based Methods

Abstract

We propose a modification that corrects for split-improvement variable importance measures in Random Forests and other tree-based methods. These methods have been shown to be biased towards increasing the importance of features with more potential splits. We show that by appropriately incorporating split-improvement as measured on out of sample data, this bias can be corrected yielding better summaries and screening tools.

add Section 3.4 to compare with other methods for dealing with similar bias; add more simulation results in Section 5; add link to Github repository for code access