Publication details
This post summarizes the peer-reviewed work cited below.
- Paper
- LeanTrace: A Resource-Conscious Lightweight Solution for Trace-Level Detection of Business Process Anomalies
- Authors
- Teoman Berkay Ayaz; Akhan Akbulut
- Presented at
- 2025 Innovations in Intelligent Systems and Applications Conference (ASYU)
- Pages
- 1–6
- Publication date
- September 10, 2025
- DOI
- 10.1109/ASYU67174.2025.11208266 — Publisher page
Detecting anomalies is crucial for maintaining operational integrity, yet state-of-the-art solutions demand heavy computational resources, which hinders their deployment in constrained environments. The paper by Teoman Berkay Ayaz (Next4biz R&D Center) and Akhan Akbulut (Next4biz R&D Center and Istanbul Kültür University) introduces LeanTrace, a new resource-conscious framework for trace-level business process anomaly detection.
LeanTrace systematically evaluates five embedding generation strategies: Word2Vec, FastText, GloVe, Node2Vec and DeepWalk, across dimensionalities from 16 to 256. These embeddings are combined with five distinct anomaly detectors: Isolation Forest, Local Outlier Factor, Elliptic Envelope, One-Class SVM, and a cluster-based detector that uses the K-Means, K-Medoids and CLARA clustering algorithms.
Tested on one public and one private dataset, LeanTrace reached up to a 0.69 F1-score (FastText + Local Outlier Factor) and up to 0.83 AUC (DeepWalk + Local Outlier Factor) on the public dataset (BPI Challenge 2020, Domestic Declarations), while the CLARA-based detector yielded recall scores of up to 0.95 on the same dataset. Running on a low-power ARM processor, LeanTrace encodes more than 10,000 traces in under a second and completes predictions in as little as 0.25 seconds. By balancing detection accuracy with a minimal footprint, the approach paves the way for deployment in resource-constrained environments.
The paper was presented at ASYU 2025 (Innovations in Intelligent Systems and Applications Conference), held in Bursa on September 10-12, 2025, and appears in the conference proceedings published by IEEE (pp. 1-6).