Research Article

Benchmarking xLSTM Architecture for Business Process Anomaly Detection

Volume: 2026 Number: 17 June 29, 2026
EN TR

Benchmarking xLSTM Architecture for Business Process Anomaly Detection

Abstract

Business Process Anomaly Detection (BPAD) is an application that can help organizations uncover deviations in their processes, which often indicate severe inefficiencies and malicious activity. In this study, we leverage a new type of Recurrent Neural Network (RNN) referred to as Extended Long-Short Term Memory (xLSTM) for the construction of unsupervised Autoencoders (AE) for use in the BPAD domain. Our study systematically compares the xLSTM architecture against three mainstream RNN architectures: Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), and Transformer, evaluating the architectural impact on anomaly detection performance. To ensure comparability, each instantiated model is constructed with an identical structure and hyperparameters, differing only in the core recurrent or self-attention blocks and evaluated with and without bidirectionality in addition to multi-head attention where applicable. The empirical results demonstrate the xLSTM- AE to be the most consistent and robust architecture across 7 datasets, yielding the highest mean F1-scores of 0.476 and 0.258, on both trace-level and event-level anomaly detection.

Keywords

Supporting Institution

the Scientific and Technological Research Council of Turkey (TUBITAK)

Project Number

TEYDEB-3231136

References

  1. [1] Ko J, Comuzzi M. A systematic review of anomaly detection for business process event logs. Business & Information Systems Engineering. 2023;65(4):441-462.
  2. [2] Musa THA, Bouras A. Prediction of next events in business processes: A deep learning approach. In: IFIP International Conference on Product Lifecycle Management; 2023. p. 210-220.
  3. [3] Ceravolo P, Damiani E, Schepis EF, Tavares GM. Real-time probing of control-flow and data-flow in event logs. Procedia Computer Science. 2022;197:751-758.
  4. [4] van Zelst SJ, Mannhardt F, de Leoni M, Koschmider A. Event abstraction in process mining: Literature review and taxonomy. Granular Computing. 2021;6:719-736.
  5. [5] Ayaz TB, Gülce E, Özcan A, Akbulut A. Semi-supervised detection of contaminated business process instances using graph autoencoders and dynamic edge convolutions for BPM anomaly detection. In: 2024 Innovations in Intelligent Systems and Applications Conference (ASYU); 2024. p. 1-6.
  6. [6] Chinnaiah V, Veerabhadram V, Aavula R, Aluvala S. PMiner: Process mining using deep autoencoder for anomaly detection and reconstruction of business processes. International Journal of Electrical and Computer Engineering Systems. 2024;15(6):531-542.
  7. [7] Hsu S, Gulce E, Ayaz TB, Ozcan A, Akbulut A. Multi-graph anomaly detection in business processes with scalable neural architectures. IEEE Access. 2025;13:34969-34984.
  8. [8] Vijayakamal M, Vasumathi D. Unsupervised learning methods for anomaly detection and log quality improvement using process event log. International Journal of Advanced Science and Technology. 2020;29(1):1109-1125.

Details

Primary Language

English

Subjects

Software Engineering (Other)

Journal Section

Research Article

Publication Date

June 29, 2026

Submission Date

June 18, 2025

Acceptance Date

November 18, 2025

Published in Issue

Year 2026 Volume: 2026 Number: 17

APA
Ayaz, T. B., Özcan, A., & Akbulut, A. (2026). Benchmarking xLSTM Architecture for Business Process Anomaly Detection. Kocaeli Journal of Science and Engineering, 2026(17), 93-103. https://doi.org/10.34088/kojose.1722112
AMA
1.Ayaz TB, Özcan A, Akbulut A. Benchmarking xLSTM Architecture for Business Process Anomaly Detection. KOJOSE. 2026;2026(17):93-103. doi:10.34088/kojose.1722112
Chicago
Ayaz, Teoman Berkay, Alper Özcan, and Akhan Akbulut. 2026. “Benchmarking XLSTM Architecture for Business Process Anomaly Detection”. Kocaeli Journal of Science and Engineering 2026 (17): 93-103. https://doi.org/10.34088/kojose.1722112.
EndNote
Ayaz TB, Özcan A, Akbulut A (June 1, 2026) Benchmarking xLSTM Architecture for Business Process Anomaly Detection. Kocaeli Journal of Science and Engineering 2026 17 93–103.
IEEE
[1]T. B. Ayaz, A. Özcan, and A. Akbulut, “Benchmarking xLSTM Architecture for Business Process Anomaly Detection”, KOJOSE, vol. 2026, no. 17, pp. 93–103, June 2026, doi: 10.34088/kojose.1722112.
ISNAD
Ayaz, Teoman Berkay - Özcan, Alper - Akbulut, Akhan. “Benchmarking XLSTM Architecture for Business Process Anomaly Detection”. Kocaeli Journal of Science and Engineering 2026/17 (June 1, 2026): 93-103. https://doi.org/10.34088/kojose.1722112.
JAMA
1.Ayaz TB, Özcan A, Akbulut A. Benchmarking xLSTM Architecture for Business Process Anomaly Detection. KOJOSE. 2026;2026:93–103.
MLA
Ayaz, Teoman Berkay, et al. “Benchmarking XLSTM Architecture for Business Process Anomaly Detection”. Kocaeli Journal of Science and Engineering, vol. 2026, no. 17, June 2026, pp. 93-103, doi:10.34088/kojose.1722112.
Vancouver
1.Teoman Berkay Ayaz, Alper Özcan, Akhan Akbulut. Benchmarking xLSTM Architecture for Business Process Anomaly Detection. KOJOSE. 2026 Jun. 1;2026(17):93-103. doi:10.34088/kojose.1722112