Work

Trustworthy Feature Importance Avoids Unrestricted Permutations

Year: 2026

Type: preprint

Abstract: Feature importance methods using unrestricted permutations are flawed due to extrapolation errors; such errors appear in all non-trivial variable importance approaches. We propose three new approaches: conditional model reliance and Knockoffs with Gaussian transformation, and restricted ALE plot designs. Theoretical and numerical results show our strategies reduce/eliminate extrapolation.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2604.11253

Authors: Borgonovo, Emanuele, Cappelli, Francesco, Lu, Xuefei, Plischke, Elmar, Rudin, Cynthia

Institutions: Bocconi University, SKEMA Business School, Clausthal University of Technology, Duke University

Topic: Explainable Artificial Intelligence (XAI)

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7154332918

Cites:

Cited by:

Related to: