Cybersecurity in intelligent transportation systems: a comparative study on AI-based anomaly detection and threat analysis

Arslan, Recep, Özseven, Turgut, Aydın, Metin Mutlu and Çelik, Yasin ORCID: https://orcid.org/0000-0002-5545-0717 2026. Cybersecurity in intelligent transportation systems: a comparative study on AI-based anomaly detection and threat analysis. Mechatronics and Intelligent Transportation Systems 5 (1) , pp. 11-30. 10.56578/mits050102

PDF - Published Version Available under License Creative Commons Attribution. Download (1MB)

Abstract

The rapid integration of technology, with increasing speeds, has transformed vehicles into cyber-physical systems by connecting them to each other Vehicle-to-Everything (V2X), significantly expanding the attack surface and leaving them vulnerable to network-based threats. Current cyber intrusion detection systems (CIDS) exhibit performance degradation due to significant class imbalance, limited resilience against adversarial attacks, and insufficient interpretability for security-critical environments. To overcome the identified issues in this study, we propose Hierarchical Classifier-Agnostic Boosted Stacking for Network Intrusion Detection (HCABS-NID), a hierarchical classifier-agnostic boosted stacking architecture for network intrusion detection in connected device ecosystems. The proposed framework adds the Synthetic Minority Over-sampling Technique for Nominal and Continuous features (SMOTENC)-based adaptive class balancing to increase minority attack detection and TreeSHAP to make it multi-level interpretable. As a hierarchical stacking strategy, a two-layer structure includes heterogeneous learners together with meta-learning, calibrated with LightGBM, XGBoost, CatBoost, and TabNet to take advantage of the complementary decision boundaries. Extensive experiments performed on the benchmark dataset from University of New South Wales Network-Based 15 (UNSW-NB15) should enhance generalization performance. HCABS-NID achieved 98.20% accuracy, 97.10% macro F1 score, and 0.989 macro Receiver Operating Characteristic Area Under the Curve (ROC-AUC), in contrast to the latest community-based methods found in the literature. The proposed model achieves 3.40 ms average inference latency, satisfying the real-time processing requirement of the V2X safety systems. Indeed, other analysis architectures show the same 96.8% accuracy at 5% corruption, which underscores their practicality. The results validate that hierarchical ensemble learning, with adaptive imbalance management artificial intelligence (AI) mechanisms, provides a sound, interpretable, and ready-to-use intelligent transportation security package.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Acadlore Publishing Services Limited
ISSN:	2958-020X
Date of First Compliant Deposit:	7 September 2026
Date of Acceptance:	27 February 2026
Last Modified:	07 Sep 2026 14:30
URI:	https://orca.cardiff.ac.uk/id/eprint/189443

Actions (repository staff only)

Edit Item