EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

PACE-IDS/IPS: Explainable and Operationally Validated Closed-Loop Intrusion Detection and Prevention for ICS/OT Networks

Rahul R, Venkatesan R, Revathi B S, Hemamalini, Anna Cruze

Computer and Decision Making: An International Journal · 31.10.2026 · 3 (6S) · 2967-2987

DOI: 10.1922/comdem.v3i6s.19046 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 31.10.2026. Verfasst von Rahul R und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Industrial control systems are responsible for keeping power plants, water plants, and other process plants running. Attacks like Stuxnet, Triton, and the 2021 intrusion into the Florida water system show that these threats are no longer rare. Yet, most studies into intrusion detection for ICS stop at developing the classifier, leaving operators with models that are reasonably accurate but difficult to deploy. In this work we build PACE-IDS/IPS, a single pipeline that detects attacks, explains them, prevents them, and learns from analyst feedback. The design follows four objectives. The first was to match published detection accuracy on at least two ICS testbeds. The second was to tag each alert with a MITRE ATT&CK for ICS technique, the affected process stage, and the asset criticality. The third was to connect the detector to an active prevention engine whose outcomes adjust the threshold without retraining. The fourth was to add the controls that any regulated site needs before going live, namely role-based access, audit logs, and IEC 62443 evidence. In the pipeline, a class-weighted XGBoost classifier is paired with an LSTM that is called only when the XGBoost score falls in an uncertain band. A small Random Forest takes the SHAP values, the active stage, and the severity as input and selects one of ten ICS techniques for each alert. The threshold is updated within hard limits by a bounded controller that uses analyst-confirmed true and false positives. We test on SWaT (1.4 M packets) and HAI 23.05 (284 K packets) using five-fold stratified cross-validation. Under cross-validation, F1 is 95.26% on SWaT and 85.00% on HAI, with FPR below 0.44%. When the starting threshold is misconfigured, the closed loop improves F1 by 40.05 points and reduces false positives by 95%, with no retraining. The learned MITRE mapper scores 95.25% on HAI and 99.44% on SWaT, which is 48 to 77 points above the rule-based baseline. P99 latency stays between 49 and 54 ms, throughput goes up to 315,000 packets per second, and 87.5% of IEC 62443 controls are met.

Zitieren

APA

R, R., R, V., S, R. B., Hemamalini, & Cruze, A. (2026). PACE-IDS/IPS: Explainable and Operationally Validated Closed-Loop Intrusion Detection and Prevention for ICS/OT Networks. Computer and Decision Making: An International Journal, 3(6S), 2967-2987. https://doi.org/10.1922/comdem.v3i6s.19046

Vancouver

R R, R V, S RB, Hemamalini, Cruze A. PACE-IDS/IPS: Explainable and Operationally Validated Closed-Loop Intrusion Detection and Prevention for ICS/OT Networks. Computer and Decision Making: An International Journal. 2026;3(6S):2967-2987. doi:10.1922/comdem.v3i6s.19046

BibTeX
@article{R2026,
  author = {R, Rahul and R, Venkatesan and S, Revathi B and Hemamalini and Cruze, Anna},
  title = {PACE-IDS/IPS: Explainable and Operationally Validated Closed-Loop Intrusion Detection and Prevention for ICS/OT Networks},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {6S},
  pages = {2967--2987},
  doi = {10.1922/comdem.v3i6s.19046}
}
Kostenlos starten →