Global Sensitivity Analysis for the Interpretation of Machine Learning Algorithms

Sonja Kuhnt, Arkadius Kalka

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

1 Scopus citations

Abstract

Global sensitivity analysis aims to quantify the importance of model input variables for a model response. We highlight the role sensitivity analysis can play in interpretable machine learning and provide a short survey on sensitivity analysis with a focus on global variance-based sensitivity measures like Sobol’ indices and Shapley values. We discuss the Monte Carlo estimation of various Sobol’ indices as well as their graphical presentation in the so-called FANOVA graphs. Global sensitivity analysis is applied to an analytical example, a Kriging model of a piston simulator and a neural net model of the resistance of yacht hulls.

Original languageEnglish
Title of host publicationArtificial Intelligence, Big Data and Data Science in Statistics
Subtitle of host publicationChallenges and Solutions in Environmetrics, the Natural Sciences and Technology
PublisherSpringer International Publishing
Pages155-169
Number of pages15
ISBN (Electronic)9783031071553
ISBN (Print)9783031071546
DOIs
StatePublished - 1 Jan 2022
Externally publishedYes

Bibliographical note

Publisher Copyright:
© The Editor (s)(if applicable) and The Author(s), under exclusive licence to Springer Nature Switzerland AG 2022.

Keywords

  • FANOVA graph
  • Global sensitivity analysis
  • Interpretable machine learning
  • Kriging
  • Shapley values
  • Sobol’ indices

Fingerprint

Dive into the research topics of 'Global Sensitivity Analysis for the Interpretation of Machine Learning Algorithms'. Together they form a unique fingerprint.

Cite this