Berti, A., Jessen, U., van der Aalst, W. M. P., & Fahland, D. (2024). Explainable Object-Centric Anomaly Detection: the Role of Domain Knowledge. In BPM-D 2024: Proceedings of the Best Dissertation Award, Doctoral Consortium, and Demonstration & Resources Forum at BPM 2024 co-located with 22nd International Conference on Business Process Management (BPM 2024) Krakow, Poland, September 1st to 6th, 2024 (blz. 162-168). (CEUR Workshop Proceedings; Vol. 3758). CEUR-WS.org. https://ceur-ws.org/Vol-3758/
Abstract
Anomaly detection is used in process mining to identify behavior differing significantly from the other instances. However, providing actionable insights out of the raw scores is challenging. In this paper, we propose three methodologies for explainable anomaly detection. In particular, we focus on object-centric event data as it increases the dimensions for anomaly detection, including the lifecycle of different objects and the interactions between them. Two of the proposed methodologies rely on the provision of domain knowledge, which can also be provided by Large Language Models (LLMs). We test the proposed techniques in a real-life case study on an (object-centric) ERP process.