An Optimized Gated Fusion Graph Neural Network Framework for Cyber-Attack Detection in Autonomous Vehicles

The rapid adoption of autonomous vehicles has significantly increased the reliance on interconnected vehicular networks, making them vulnerable to diverse cyber-attacks that can compromise safety, reliability, and real-time decision-making. Existing cyber-attack detection methods often suffer from limited adaptability to evolving attack patterns, high false positive rates, and increased computational complexity, which restrict their applicability in real-world autonomous vehicle environments. To address these challenges, this paper proposes an Optimized Gated Fusion Adaptive Graph Neural Network for Cyber-Attack Detection on Autonomous Vehicles (GFAGNN-CD-AV). The proposed framework employs a Noise-Tolerant-Matched Filter (NTMF) for robust data preprocessing and an Adaptive Elite Ant Lion Optimizer (AEALO) to select discriminative network traffic features. A Gated Fusion Adaptive Graph Neural Network is then used to effectively model complex relationships in vehicular network data, while the Lotus Effect Optimization Algorithm (LEOA) optimizes model parameters to enhance detection accuracy and reduce false positives. The proposed method is evaluated using the VeReMi dataset under both normal and attack scenarios. Experimental results demonstrate that GFAGNN-CD-AV achieves superior performance with an accuracy of 99.85%, a precision of 98.00%, a recall of 97.00%, and a reduced detection latency of 65 ms, outperforming existing state-of-the-art cyber-attack detection approaches. These results confirm the effectiveness and suitability of the proposed framework for real-time cybersecurity in autonomous vehicle networks.