Debiasing SHAP scores in random forests

Abstract Black box machine learning models are currently being used for high-stakes decision making in various parts of society such as healthcare and criminal justice. While tree-based ensemble methods random forests typically outperform deep on tabular data sets, their built-in variable importance algorithms known to be strongly biased toward high-entropy features. It was recently shown that the increasingly popular SHAP (SHapley Additive exPlanations) values suffer from a similar bias. We propose debiased or "shrunk" scores based sample splitting which additionally enable detection overfitting issues at feature level.