Trustworthy Feature Importance Avoids Unrestricted Permutations

TL;DR

This paper identifies flaws in traditional feature importance methods using unrestricted permutations and introduces three new approaches that reduce or eliminate extrapolation errors, supported by theoretical and numerical evidence.

Contribution

It proposes three novel feature importance techniques—conditional model reliance, Knockoffs with Gaussian transformation, and restricted ALE plots—that address extrapolation issues in existing methods.

Findings

Traditional permutation-based importance methods suffer from extrapolation errors.

The proposed methods effectively reduce or eliminate extrapolation in feature importance.

Theoretical and numerical results validate the effectiveness of the new approaches.

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.