Evaluation of Visualization Concepts for Explainable Machine Learning Methods in the Context of Manufacturing

Alexander Gerling, Alexander Gerling, Alexander Gerling, Christian Seiffer, Holger Ziekow, Ulf Schreier, Andreas Hess, Djaffar Abdeslam, Djaffar Abdeslam

2021

Abstract

Machine Learning (ML) is increasingly used in the manufacturing domain to identify the root cause of product errors. A product error can be difficult to identify and most ML models are not easy to understand. Therefore, we investigated visualization techniques for use in manufacturing. We conducted several interviews with quality engineers and a group of students to determine the usefulness of 15 different visualizations. These are mostly state-of-the-art visualizations or adjusted visualizations for our use case. The objective is to prevent misinterpretations of results and to help making decisions more quickly. The most popular visualizations were the Surrogate Decision Tree Model and the Scatter Plot because they show simple illustrations that are easy to understand. We also discuss eight combinations of visualizations to better identify the root cause of an error.

Paper Citation

in Harvard Style

Gerling A., Seiffer C., Ziekow H., Schreier U., Hess A. and Abdeslam D. (2021). Evaluation of Visualization Concepts for Explainable Machine Learning Methods in the Context of Manufacturing. In Proceedings of the 5th International Conference on Computer-Human Interaction Research and Applications - Volume 1: CHIRA, ISBN 978-989-758-538-8, pages 189-201. DOI: 10.5220/0010688900003060

in Bibtex Style

@conference{chira21,
author={Alexander Gerling and Christian Seiffer and Holger Ziekow and Ulf Schreier and Andreas Hess and Djaffar Abdeslam},
title={Evaluation of Visualization Concepts for Explainable Machine Learning Methods in the Context of Manufacturing},
booktitle={Proceedings of the 5th International Conference on Computer-Human Interaction Research and Applications - Volume 1: CHIRA,},
year={2021},
pages={189-201},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010688900003060},
isbn={978-989-758-538-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 5th International Conference on Computer-Human Interaction Research and Applications - Volume 1: CHIRA,
TI - Evaluation of Visualization Concepts for Explainable Machine Learning Methods in the Context of Manufacturing
SN - 978-989-758-538-8
AU - Gerling A.
AU - Seiffer C.
AU - Ziekow H.
AU - Schreier U.
AU - Hess A.
AU - Abdeslam D.
PY - 2021
SP - 189
EP - 201
DO - 10.5220/0010688900003060