EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Attention-Based Transformer Architectures for Cyber Physical System Threat Detection

Stephanie Ness, Subhash Lakshminarayana

IEEE Access · 2026 · 14 · 111798-111814

DOI: 10.1109/access.2026.3715110 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Stephanie Ness und 1 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Cyber-Physical Systems (CPSs) are now at the heart of many industrial environments. Still, as their scale and integration increase, they also become vulnerable to a wide variety of cyber and operational threat actors. In this paper, we evaluated four Transformer-based models (TabTransformer, FTTransformer, Residual Transformer, and an Attention-Pooling Transformer) on two distinct datasets: the TON_IoT Network Dataset and the CyberFedDefender Threat Detection Dataset. The models used column-level embeddings and attention layers in order to model dependencies across mixed tabular features, which can be especially difficult to model with more classical approaches. TabTransformer outperformed models in both datasets. On the TON_IoT Network Dataset, it achieved an outstanding performance of 99.91% accuracy and AUC of 0.9947. Performance degraded on the CyberFedDefender dataset with its greater variety of attack behaviors; nonetheless, TabTransformer models yielded the highest accuracy (48.95%), thus demonstrating a superior capacity to detect less prevalent or more complex detections. Overall, our findings suggest that attention-based Transformers are effective for threat detection, in particular when feature interactions are important in a CPS setting.

Zitieren

APA

Ness, S., & Lakshminarayana, S. (2026). Attention-Based Transformer Architectures for Cyber Physical System Threat Detection. IEEE Access, 14, 111798-111814. https://doi.org/10.1109/access.2026.3715110

Vancouver

Ness S, Lakshminarayana S. Attention-Based Transformer Architectures for Cyber Physical System Threat Detection. IEEE Access. 2026;14:111798-111814. doi:10.1109/access.2026.3715110

BibTeX
@article{Ness2026,
  author = {Ness, Stephanie and Lakshminarayana, Subhash},
  title = {Attention-Based Transformer Architectures for Cyber Physical System Threat Detection},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {111798--111814},
  doi = {10.1109/access.2026.3715110}
}
Kostenlos starten →