VARSHAP: Addressing Global Dependency Problems in Explainable AI with Variance-Based Local Feature Attribution

TL;DR

VARSHAP is a new local feature attribution method that uses variance reduction to improve explainability in AI models, especially under data distribution shifts.

Contribution

It introduces VARSHAP, a model-agnostic attribution method based on variance reduction, which satisfies Shapley axioms and is resilient to global data shifts.

Findings

Outperforms KernelSHAP and LIME in experiments

Resilient to global data distribution shifts

Demonstrates effectiveness on synthetic and real datasets

Abstract

Existing feature attribution methods like SHAP often suffer from global dependence, failing to capture true local model behavior. This paper introduces VARSHAP, a novel model-agnostic local feature attribution method which uses the reduction of prediction variance as the key importance metric of features. Building upon Shapley value framework, VARSHAP satisfies the key Shapley axioms, but, unlike SHAP, is resilient to global data distribution shifts. Experiments on synthetic and real-world datasets demonstrate that VARSHAP outperforms popular methods such as KernelSHAP or LIME, both quantitatively and qualitatively.