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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gajewski, Mateusz, Morzy, Mikołaj, Karczmarz, Adam, Sankowski, Piotr
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915333157158912
author Gajewski, Mateusz
Morzy, Mikołaj
Karczmarz, Adam
Sankowski, Piotr
author_facet Gajewski, Mateusz
Morzy, Mikołaj
Karczmarz, Adam
Sankowski, Piotr
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_07229
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VARSHAP: Addressing Global Dependency Problems in Explainable AI with Variance-Based Local Feature Attribution
Gajewski, Mateusz
Morzy, Mikołaj
Karczmarz, Adam
Sankowski, Piotr
Machine Learning
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.
title VARSHAP: Addressing Global Dependency Problems in Explainable AI with Variance-Based Local Feature Attribution
topic Machine Learning
url https://arxiv.org/abs/2506.07229