A Guide to Feature Importance Methods for Scientific Inference

Fiona Katharina Ewald, Ludwig Bothmann, Marvin N. Wright, Bernd Bischl, Giuseppe Casalicchio*, Gunnar König

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

1 Citation (Scopus)
1 Downloads (Pure)

Abstract

While machine learning (ML) models are increasingly used due to their high predictive power, their use in understanding the data-generating process (DGP) is limited. Understanding the DGP requires insights into feature-target associations, which many ML models cannot directly provide due to their opaque internal mechanisms. Feature importance (FI) methods provide useful insights into the DGP under certain conditions. Since the results of different FI methods have different interpretations, selecting the correct FI method for a concrete use case is crucial and still requires expert knowledge. This paper serves as a comprehensive guide to help understand the different interpretations of global FI methods. Through an extensive review of FI methods and providing new proofs regarding their interpretation, we facilitate a thorough understanding of these methods and formulate concrete recommendations for scientific inference. We conclude by discussing options for FI uncertainty estimation and point to directions for future research aiming at full statistical inference from black-box ML models.

Original languageEnglish
Title of host publicationExplainable Artificial Intelligence - Second World Conference, xAI 2024, Proceedings
EditorsLuca Longo, Sebastian Lapuschkin, Christin Seifert
Number of pages25
PublisherSpringer
Publication date2024
Pages440-464
ISBN (Print)9783031637964
DOIs
Publication statusPublished - 2024
Event2nd World Conference on Explainable Artificial Intelligence, xAI 2024 - Valletta, Malta
Duration: 17 Jul 202419 Jul 2024

Conference

Conference2nd World Conference on Explainable Artificial Intelligence, xAI 2024
Country/TerritoryMalta
CityValletta
Period17/07/202419/07/2024
SeriesCommunications in Computer and Information Science
Volume2154 CCIS
ISSN1865-0929

Bibliographical note

Funding Information:
MNW was supported by the German Research Foundation (DFG), Grant Numbers: 437611051, 459360854. GK was supported by the German Research Foundation through the Cluster of Excellence \u201CMachine Learning - New Perspectives for Science\u201D (EXC 2064/1 number 390727645).

Publisher Copyright:
© The Author(s) 2024.

Keywords

  • Feature Importance
  • Interpretable ML
  • Model-agnostic Interpretability

Cite this