Scholarlink

<p>The widespread adoption of Internet of Things (IoT) technologies has transformed intelligent transportation systems (ITS) into a fundamental pillar of smart city infrastructures, enabling real-time traffic monitoring, connected vehicle communication, and adaptive mobility management over large-scale distributed networks. Despite these advancements, IoT-enabled transportation systems remain highly vulnerable to sophisticated cyber-attacks due to their decentralized architecture, heterogeneous devices, continuous data exchange, and stringent latency constraints. Existing cybersecurity solutions predominantly focus on reactive intrusion detection and post-incident mitigation, which are insufficient for mission-critical transportation services where even short disruptions can lead to severe safety and operational consequences. To address this challenge, this paper proposes a proactive deep learning–based cyber-attack avoidance framework tailored for IoT-enabled smart city transportation networks. The proposed approach models the transportation communication infrastructure as a dynamic graph and employs a hybrid spatio-temporal Graph Transformer architecture to jointly capture network topology–aware attack propagation patterns and long-range temporal dependencies in evolving cyber threats. A self-supervised pretraining strategy based on masked traffic modeling is incorporated to enhance robustness against zero-day and previously unseen attacks. Furthermore, a risk-aware decision layer is introduced to translate predicted attack trajectories into adaptive avoidance actions, including dynamic access control, intelligent routing adjustments, and selective node isolation, thereby preventing attack escalation before service disruption occurs. Experimental evaluation across UNSW-NB15, CICIDS2018, Bot-IoT, and Smart-ITS datasets demonstrate that the proposed framework consistently outperforms state-of-the-art models, achieving up to 98.64% accuracy, 98.04% F1-score, macro-F1 of 0.964, false alarm rates below 1%, and real-time response latency as low as 34.9&#xa0;ms, while maintaining strong zero-day detection capability (AUROC = 0.963). The proposed framework delivers accurate, low-latency, and robust intrusion detection across diverse and large-scale cyber-physical environments, demonstrating its suitability for real-time deployment in next-generation intelligent networks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A deep learning–driven cyber attack avoidance framework for secure IoT-enabled smart city transportation systems

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Prashant Kumar Shukla,
  • Ratish Agarwal

摘要

The widespread adoption of Internet of Things (IoT) technologies has transformed intelligent transportation systems (ITS) into a fundamental pillar of smart city infrastructures, enabling real-time traffic monitoring, connected vehicle communication, and adaptive mobility management over large-scale distributed networks. Despite these advancements, IoT-enabled transportation systems remain highly vulnerable to sophisticated cyber-attacks due to their decentralized architecture, heterogeneous devices, continuous data exchange, and stringent latency constraints. Existing cybersecurity solutions predominantly focus on reactive intrusion detection and post-incident mitigation, which are insufficient for mission-critical transportation services where even short disruptions can lead to severe safety and operational consequences. To address this challenge, this paper proposes a proactive deep learning–based cyber-attack avoidance framework tailored for IoT-enabled smart city transportation networks. The proposed approach models the transportation communication infrastructure as a dynamic graph and employs a hybrid spatio-temporal Graph Transformer architecture to jointly capture network topology–aware attack propagation patterns and long-range temporal dependencies in evolving cyber threats. A self-supervised pretraining strategy based on masked traffic modeling is incorporated to enhance robustness against zero-day and previously unseen attacks. Furthermore, a risk-aware decision layer is introduced to translate predicted attack trajectories into adaptive avoidance actions, including dynamic access control, intelligent routing adjustments, and selective node isolation, thereby preventing attack escalation before service disruption occurs. Experimental evaluation across UNSW-NB15, CICIDS2018, Bot-IoT, and Smart-ITS datasets demonstrate that the proposed framework consistently outperforms state-of-the-art models, achieving up to 98.64% accuracy, 98.04% F1-score, macro-F1 of 0.964, false alarm rates below 1%, and real-time response latency as low as 34.9 ms, while maintaining strong zero-day detection capability (AUROC = 0.963). The proposed framework delivers accurate, low-latency, and robust intrusion detection across diverse and large-scale cyber-physical environments, demonstrating its suitability for real-time deployment in next-generation intelligent networks.