Improving KernelSHAP: Practical Shapley Value Estimation via Linear Regression

TL;DR

This paper improves the efficiency and reliability of KernelSHAP, a popular method for estimating Shapley values in machine learning model explanations, by analyzing and enhancing linear regression-based estimators.

Contribution

It introduces unbiased and variance-reduction techniques for KernelSHAP, along with methods to detect convergence and estimate uncertainty, advancing practical Shapley value estimation.

Findings

Original KernelSHAP has negligible bias increase with lower variance.

Variance reduction accelerates convergence of Shapley value estimators.

New stochastic game-based KernelSHAP offers faster global explanations.

Abstract

The Shapley value concept from cooperative game theory has become a popular technique for interpreting ML models, but efficiently estimating these values remains challenging, particularly in the model-agnostic setting. Here, we revisit the idea of estimating Shapley values via linear regression to understand and improve upon this approach. By analyzing the original KernelSHAP alongside a newly proposed unbiased version, we develop techniques to detect its convergence and calculate uncertainty estimates. We also find that the original version incurs a negligible increase in bias in exchange for significantly lower variance, and we propose a variance reduction technique that further accelerates the convergence of both estimators. Finally, we develop a version of KernelSHAP for stochastic cooperative games that yields fast new estimators for two global explanation methods.