SPECTRE: Decontamination and Anomaly Detection on Business Process Event Logs

The study titled “SPECTRE: A Multi-Purpose Autoencoder Model for Decontamination and Anomaly Detection on Anomalous Business Process Event Logs”, prepared by Next4biz R&D Center researchers in academic collaboration, was published on July 14, 2026 in Cluster Computing: The Journal of Networks, Software Tools and Applications, an international peer-reviewed journal published by Springer Nature. The article was authored by Teoman Berkay Ayaz, Alper Özcan, and Akhan Akbulut.

The Journal

Covering current research in parallel and distributed systems, artificial intelligence, cloud computing, software, and high-performance computing, Cluster Computing has an impact factor of 5.5 and is indexed by SCIE and Scopus. In the JCR assessment, the journal is ranked Q1 in the Computer Science, Information Systems and Computer Science, Theory & Methods categories — in other words, in the top 25% of scientific journals in its field.

Our Study: SPECTRE

Introduced in the article, SPECTRE (Spectral-channel Parallel Encoder with Contextual Attention for Reconstruction and Enhancement) is an unsupervised, multi-purpose autoencoder model developed to clean erroneous and anomalous data in event logs — the digital footprints of business processes — and to detect process anomalies. Since event logs are digital reflections of real-world operational flows, deviations ranging from simple operational mishaps to fraud are captured in these records and can lead to significant losses when left undetected.

Tested on a total of 11 datasets — 7 public and 4 private — the model achieved up to 99.94% accuracy in decontamination and an F1 score of 0.996 in event-level anomaly detection, while preserving near-linear scalability on high volumes of process data. The study also employed a data augmentation method called Reduction Window to strengthen the model on datasets with scarce traces; this approach yielded up to a 2% improvement in decontamination accuracy and up to a 45% gain in anomaly detection performance on challenging logs.

You can access the article on Springer: SPECTRE: A Multi-Purpose Autoencoder Model for Decontamination and Anomaly Detection on Anomalous Business Process Event Logs (DOI: 10.1007/s10586-026-06258-8)