Home // DATA ANALYTICS 2023, The Twelfth International Conference on Data Analytics // View article
Authors:
Holger Ziekow
Peter Schanbacher
Keywords: Interpretability, Understandability; Explainability; explainable AI; XAI; human-centered AI; black-box models.
Abstract:
XAI methods, such as partial dependency plots or individual conditional expectation plots help understanding the impact of feature values on the output of an AI model. However, these techniques can only analyze the concepts manifested in a single feature. This makes it hard to investigate the impact of higher-level concepts, spanning across multiple features (E.g. a model prediction may depend on the morbidity of a patient, while morbidity is only indirectly reflected through features about symptoms). In this paper, we present and test a concept for getting insight into model dependency on aspects on a higher semantic level. This enables an understanding how a model output changes in dependence on meaningful higher-level concepts and aids data scientists in analyzing machine learning models.
Pages: 1 to 6
Copyright: Copyright (c) IARIA, 2023
Publication date: September 25, 2023
Published in: conference
ISSN: 2308-4464
ISBN: 978-1-68558-111-4
Location: Porto, Portugal
Dates: from September 25, 2023 to September 29, 2023