Publication details
This post summarizes the peer-reviewed work cited below.
- Paper
- Graph Based Business Process Anomaly Detection with Edge Feature Reconstruction and Advanced Linear Networks
- Authors
- Teoman Berkay Ayaz; Rabia Çevik; Alper Özcan; Akhan Akbulut
- Presented at
- 2025 7th International Congress on Human-Computer Interaction, Optimization and Robotic Applications (ICHORA)
- Pages
- 1–6
- Publication date
- May 23, 2025
- DOI
- 10.1109/ichora65333.2025.11017131 — Publisher page
- Authors' version (PDF)
© 2025 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses. This is the authors' version; the published version is available at doi.org/10.1109/ichora65333.2025.11017131.
Business Process Management (BPM) is a critical field for keeping business operations running soundly; spotting inefficiencies and potentially malicious activity in processes early matters even more as the pace of business accelerates. This need gave rise to the research area known as "business process anomaly detection." This paper, prepared by the Next4biz R&D Center together with Akdeniz University and Istanbul Kültür University, takes the team's earlier work on graph-based anomaly detection one step further.
In the study, process data is represented as a graph and a Graph Autoencoder (GAE) architecture built with different graph convolution operators is designed; the goal is to raise performance through the reconstruction of edge features and advanced linear networks. The authors systematically compare 3 different encoder architectures and 4 different decoder options across six separate datasets. One of the datasets consists of records from the Next4biz BPM platform (6,178 traces, 44,413 events); the other five are the event logs on travel expense claims published as part of the BPI Challenge 2020.
The results differ markedly depending on the encoder-decoder combination: in anomaly detection the F1 score ranges from 0.219 to 0.674. This wide range clearly shows how much the architectural choices affect the outcome.
The paper was presented at the 7th International Congress on Human-Computer Interaction, Optimization and Robotic Applications (ICHORA 2025), held in Ankara on May 23–24, 2025, and was published by IEEE in the congress proceedings (pp. 1–6). The authors are Teoman Berkay Ayaz and Rabia Çevik (Next4biz R&D Center), Alper Özcan (Akdeniz University, Computer Engineering) and Akhan Akbulut (Next4biz R&D Center and Istanbul Kültür University).