CoChat

AI-Based Anomaly Detection in Industrial Control and Cyber–Physical Systems: A Data-Type-Oriented Systematic Review

Open access
Seo, Jung Kyu, Lee, JuHyeon, Kim, Buyoung, Shim, Wooseong, Seo, Jung Taek · 2025 · Electronics

The DOI matches its Crossref record; 132 of the 132 references it cites were found in a registry or an index; 149 of the 461 sources its claims cite were checked for support.

Abstract

A systematic review of anomaly detection in industrial control and cyber-physical systems that highlights benchmark concentration, label sparsity, concept drift, and insufficient consideration of real-time resource-constrained environments.

Overview

CoChat Summary

TL;DR

This PRISMA-based systematic review analyzes 89 primary studies (2021–2025) on AI-based anomaly detection in ICS/CPS, organizing them by five data modalities: network traffic, process, simulation, hybrid, and other auxiliary data. It finds complementary strengths across modalities but persistent limitations in benchmark diversity, zero-day generalization, label sparsity, concept drift, and real-time OT deployment.

Key findings

  • The review included 89 studies; publication counts by year were 2021: 12, 2022: 16, 2023: 15, 2024: 23, and 2025: 23; by reference type, 68 journal articles and 21 conference proceedings.
  • Network traffic data mainly supports cyber-layer monitoring (e.g., reconnaissance, DoS, MITM), while process data targets physical-layer anomalies (e.g., process disturbances, FDI, stealth deviations); simulation and hybrid data support rare-scenario generation and cyber-physical consistency checking.
  • Reported performance examples include ADT achieving F1 scores of 0.995–0.999 on SWaT, WADI, and HAI; Li et al. reporting average F1 above 91% across seven CPS datasets; Wang et al. reporting approximately 90% detection accuracy with 2–4 min average detection delay; and Kus et al. finding recall for unseen attack families as low as 3.2–14.7%.
  • Future directions identified are multi-domain dataset development, physics- and control-informed model design, hybrid-data-driven integrated detection, and lightweight edge deployment.

Methods

  • PRISMA 2020 systematic literature review; the review protocol was not registered. Searches covered Web of Science, Scopus, IEEE Xplore, and the ACM Digital Library, plus a Google Scholar manual search, for studies published 1 January 2021–31 October 2025.
  • Eligibility required English-language, peer-reviewed conference or journal studies presenting an anomaly detection technique based on a specific dataset type and addressing cyber-attack anomaly detection for ICS/CPS; exclusions included studies without objective evaluation, device-fault or physical-behavior anomalies, secondary research, and general IT-only studies.
  • Screening began with 225 records; 22 duplicates and 6 out-of-period records were removed, 40 were excluded at title/abstract screening, 37 at full-text evaluation, 120 underwent quality assessment, 31 were excluded for insufficient methodological details, and 89 were included. Data extraction used an EBSE-based form; two authors extracted data and the remaining authors cross-validated.

Limitations

  • Stated: heavy reliance on a small number of benchmark datasets and narrowly scoped testbeds; scarcity of long-term, multi-domain datasets; label sparsity and concept drift; and evaluation methodologies that insufficiently account for real-world OT operating conditions and resource constraints.
  • Stated: the review protocol was not registered in any database.
  • Stated in the PRISMA checklist: no statistical meta-analysis, no formal reporting-bias assessment, and certainty of evidence was not formally graded; risk of bias was not specifically tabulated, only described generally.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/electronics15010020

Cites

132 of 132 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

100 Valid · 7 Not valid · 118 Not checked · 132 Own results and interpretations

179 more claims · 56 with their cited sources checked

Create a free account to see all 357 claims and to verify the 1 reference still open.

See all claims — free account
  • Probably not a substantial claim. 69
  • It may state two separate claims as one. 129

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 12 · the rest aren’t loaded

7 more findings

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Show all 132 references

Research works printed but not linked: 18

  1. Abshari, D., and Sridhar, M. (2025). A survey of anomaly detection in cyber-physical systems. arXiv.
  2. Djouad, A., Atil, F., Seriai, A.-D., and Beddiar, C. (2018, January 1–2). Domain Model for Cyber-Physical Systems. Proceedings of the ICAASE, Constantine, Algeria.
  3. Keele, S. (2007). Guidelines for Performing Systematic Literature Reviews in Software Engineering; Technical Report, Ver. 2.3 ebse Technical Report. ebse: 2007, Elsevier.
  4. Sangeetha, V., Naidu, R.C.A., Bhat, A., and Kulkarni, P. (2024, January 13–14). Integrating deep learning with ensemble approach for anomaly detection in network traffic. Proceedings of the 2024 4th International Conference on Mobile Networks and Wireless Communications (ICMNWC), Reykjavik, Iceland.
  5. Zhang, 2024, Unsupervised Real-time Communication Traffic Anomaly Detection for Multi-dimensional Industrial Networks, IEEE Trans. Ind. Cyber-Phys. Syst., 3, 228
  6. Moustafa, N. (2019, January 21–25). New generations of internet of things datasets for cybersecurity applications based machine learning: TON_IoT datasets. Proceedings of the eResearch Australasia Conference, Brisbane, Australia.
  7. Frazão, I., Abreu, P., Cruz, T., Araújo, H., and Simões, P. (IEEE Dataport, 2019). Cyber-Security Modbus ICS Dataset, IEEE Dataport.
  8. Morris, T.H., Thornton, Z., and Turnipseed, I. (2015, January 3–4). Industrial control system simulation and data logging for intrusion detection system research. Proceedings of the 7th Annual Southeastern Cyber Security Summit, Huntsvile, AL, USA. Available online: https://www.semanticscholar.org/paper/Industrial-Control-System-Simulation-and-Data-for-Morris-Thornton/bb9714e0c661576f5df19fb54e0e26567ca37372.
  9. Ring, M., Wunderlich, S., Grüdl, D., Landes, D., and Hotho, A. (2017, January 29–30). Flow-based benchmark data sets for intrusion detection. Proceedings of the 16th European Conference on Cyber Warfare and Security (ECCWS), Dublin, Ireland.
  10. Cho, K., Mitsuya, K., and Kato, A. (2000, January 18–23). Traffic data repository at the {WIDE} project. Proceedings of the 2000 USENIX Annual Technical Conference (USENIX ATC 00), San Diego, CA, USA.
  11. Soltani, M., Knudsen, T., and Bak, T. (2009, January 14–16). Modeling and simulation of offshore wind farms for farm level control. Proceedings of the European Offshore Wind Conference and Exhibition (EOW), Stockholm, Sweden.
  12. Shin, H.-K., Lee, W., Yun, J.-H., and Kim, H. (2020, January 10). {HAI} 1.0:{HIL-based} augmented {ICS} security dataset. Proceedings of the 13th USENIX Workshop on Cyber Security Experimentation and Test (CSET 20), Online.
  13. Ayas, 2025, Detecting cyberattacks based on deep neural network approaches in industrial control systems, J. Inf. Secur. Appl., 94, 104206
  14. Formby, D., Rad, M., and Beyah, R. (2018, January 13). Lowering the barriers to industrial control system security with {GRFICS}. Proceedings of the 2018 USENIX Workshop on Advances in Security Education (ASE 18), Baltimore, MD, USA.
  15. Li, 2023, Functional Pattern-Related Anomaly Detection Approach Collaborating Binary Segmentation with Finite State Machine, Comput. Mater. Contin., 77, 3573
  16. MR, 2025, Assessing the Effectiveness of PCAT in Avoiding Process Anomalies in Water Treatment Plants, IEEE Trans. Ind. Inform., 99, 1
  17. Rieth, 2017, Additional tennessee eastman process simulation data for anomaly detection evaluation, Harv. Dataverse, 1, 2017
  18. (2020). IEEE Standard for Information Technology—Telecommunications and Information Exchange Between Systems—Local and Metropolitan Area Networks—Specific Requirements—Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Institute of Electrical and Electronics Engineers (IEEE).

Cited by

See all 15

Metadata

Authors
Seo, J. K., Lee, J., Kim, B., Shim, W., & Seo, J. T.
Affiliations
  • Gachon University (KR)
  • Korea Research Institute of Ships and Ocean Engineering (KR)
  • Korea Institute of Ships & Ocean Engineering, Daejeon 34103, Republic of Korea
Corresponding author
Seo, Jung Taek
Venue
Electronics
Publisher
MDPI AG
Published
2025-12-20
Volume
15
Issue
1
Pages
20
ISSN
2079-9292 (ISSN-L)
OpenAlex
W7116652254
Cited by
15
References
115
Citations by year
2026: 15
FWCI
2.19
Type
journal article
Language
en
Open access
Locations
Electronics (publishedVersion, cc-by)
Funders
Institute of Information & communications Technology Planning & Evaluation (RS-2023-00241376)
Topics
  • Physical Sciences › Computer Science › Computer Networks and Communications › Network Security and Intrusion Detection
  • Physical Sciences › Engineering › Control and Systems Engineering › Smart Grid Security and Resilience
  • Physical Sciences › Engineering › Control and Systems Engineering › Fault Detection and Control Systems
Keywords
anomaly detection, industrial control systems, cyber-physical systems, critical infrastructure security, cyberattacks, network traffic data, operational data, simulated data, hybrid data, false data injection attacks, stealthy attacks, denial-of-service attacks, man-in-the-middle attacks, reconnaissance attacks, edge deployment, concept drift, label sparsity

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.