EUVIMEDEUVIMED entdecken
Crossref

Auditing Process Knowledge for Parsimonious and Explainable Industrial Cyberattack Detection

Hyunwook Kim, Keith Wilkinson, Renjith Prasad Kaippilly Mana, Aahan Rathod, Chathurangi Shyalika, Amit Sheth

Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 633-640

DOI: 10.1609/aaaiss.v10i1.43466 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Hyunwook Kim und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Industrial Control Systems (ICSs) require attack detection that captures temporal behavior while remaining interpretable to operators. Engineering-informed detectors often incorporate physically plausible process rules, but physical plausibility alone does not ensure that a relationship is reliable or useful for detection: it may be weakly supported by observed data, unstable across operating conditions, or redundant with other process evidence. We therefore study a different question: which engineering relationships should actually be admitted into a detector? Using the Secure Water Treatment (SWaT) dataset, we audit 27 candidate process relationships against training and validation data. Six relationships that pass all audit checks form a conservative CORE representation, while 16 are retained overall. We combine CORE with a 25-variable LSTM and evaluate it on a fixed chronological test partition across five random seeds. CORE achieves the highest mean test MCC (0.247) and F1 (0.161), compared with 0.183 and 0.137 for the LSTM alone. Using all 16 retained relationships performs worse than CORE, suggesting that more process knowledge does not necessarily improve detection. However, all models perform substantially worse on the chronological test data than on validation data, revealing a major limitation in generalization over time. Our contribution is a systematic way to audit engineering knowledge before incorporating it into a hybrid detector, favoring a smaller and more interpretable representation.

Zitieren

APA

Kim, H., Wilkinson, K., Mana, R. P. K., Rathod, A., Shyalika, C., & Sheth, A. (2026). Auditing Process Knowledge for Parsimonious and Explainable Industrial Cyberattack Detection. Proceedings of the AAAI Symposium Series, 10(1), 633-640. https://doi.org/10.1609/aaaiss.v10i1.43466

Vancouver

Kim H, Wilkinson K, Mana RPK, Rathod A, Shyalika C, Sheth A. Auditing Process Knowledge for Parsimonious and Explainable Industrial Cyberattack Detection. Proceedings of the AAAI Symposium Series. 2026;10(1):633-640. doi:10.1609/aaaiss.v10i1.43466

BibTeX
@article{Kim2026,
  author = {Kim, Hyunwook and Wilkinson, Keith and Mana, Renjith Prasad Kaippilly and Rathod, Aahan and Shyalika, Chathurangi and Sheth, Amit},
  title = {Auditing Process Knowledge for Parsimonious and Explainable Industrial Cyberattack Detection},
  journal = {Proceedings of the AAAI Symposium Series},
  year = {2026},
  volume = {10},
  number = {1},
  pages = {633--640},
  doi = {10.1609/aaaiss.v10i1.43466}
}
Kostenlos starten →