Home // International Journal On Advances in Software, volume 17, numbers 1 and 2, 2024 // View article

Regional Feature Importance for Error Analysis in Manufacturing

Authors:
Valentin Göttisheim
Holger Ziekow
Peter Schanbacher

Keywords: eXplainable Artificial Intelligence; XAI; SHapley Additive exPlanations; SHAP; feature importance, manufacturing quality management; error analysis.

Abstract:
Quality management in manufacturing can benefit from integration of artificial intelligence to detect and analyze errors in production. However, finding the causes of errors requires not only accurate predictions, but also suitable explanations of the underlying data analysis processes. This paper extends our previous work on a novel approach to measure the importance of features for error analysis. This approach bridges the gap between global and local importance and introduces the concept of regional feature importance, which captures the impact of features in specific regions of the feature space, rather than globally or locally. We generalize this method as a task of partitioning the feature space and aggregating the local importance of features within these regions. Our findings demonstrate that this approach can reveal interesting and actionable insights for quality management in manufacturing.

Pages: 100 to 115

Copyright: Copyright (c) to authors, 2024. Used with permission.

Publication date: June 30, 2024

Published in: journal

ISSN: 1942-2628