Trustworthy Feature Importance Avoids Unrestricted Permutations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Borgonovo, Emanuele, Cappelli, Francesco, Lu, Xuefei, Plischke, Elmar, Rudin, Cynthia
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913025896742912
author Borgonovo, Emanuele
Cappelli, Francesco
Lu, Xuefei
Plischke, Elmar
Rudin, Cynthia
author_facet Borgonovo, Emanuele
Cappelli, Francesco
Lu, Xuefei
Plischke, Elmar
Rudin, Cynthia
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_11253
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Trustworthy Feature Importance Avoids Unrestricted Permutations
Borgonovo, Emanuele
Cappelli, Francesco
Lu, Xuefei
Plischke, Elmar
Rudin, Cynthia
Machine Learning
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.
title Trustworthy Feature Importance Avoids Unrestricted Permutations
topic Machine Learning
url https://arxiv.org/abs/2604.11253