A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values

TL;DR

This paper introduces a unified framework for estimating Shapley values with provable efficiency, providing theoretical guarantees and practical improvements for model explanations in machine learning.

Contribution

It presents the first unified theoretical framework for various Shapley value estimators, including KernelSHAP, with non-asymptotic guarantees and scalable implementation strategies.

Findings

The framework offers strong theoretical bounds for estimator performance.

Empirical results show low error with modest sample sizes.

Scalable methods outperform existing KernelSHAP implementations on large datasets.

Abstract

Shapley values have emerged as a critical tool for explaining which features impact the decisions made by machine learning models. However, computing exact Shapley values is difficult, generally requiring an exponential (in the feature dimension) number of model evaluations. To address this, many model-agnostic randomized estimators have been developed, the most influential and widely used being the KernelSHAP method (Lundberg & Lee, 2017). While related estimators such as unbiased KernelSHAP (Covert & Lee, 2021) and LeverageSHAP (Musco & Witter, 2025) are known to satisfy theoretical guarantees, bounds for KernelSHAP have remained elusive. We describe a broad and unified framework that encompasses KernelSHAP and related estimators constructed using both with and without replacement sampling strategies. We then prove strong non-asymptotic theoretical guarantees that apply to all estimators from our framework. This provides, to the best of our knowledge, the first theoretical guarantees for KernelSHAP and sheds further light on tradeoffs between existing estimators. Through comprehensive benchmarking on small and medium dimensional datasets for Decision-Tree models, we validate our approach against exact Shapley values, consistently achieving low mean squared error with modest sample sizes. Furthermore, we make specific implementation improvements to enable scalability of our methods to high-dimensional datasets. Our methods, tested on datasets such MNIST and CIFAR10, provide consistently better results compared to the KernelSHAP library.