Controller Area Network Flexible Data Attack Detection Using Deep Learning Mechanism

<p>CAN-FD bus intrusion detection in autonomous systems must be performed in real time with minimal false<br>positives. We propose an attention-based CNN-LSTM hybrid method that combines spatial feature ex-<br>traction (CNN) and temporal anomaly detection (LSTM with attention). We tested our approach in the<br>CAN-FD-Attack-Sim dataset and achieved an F1 score of 97.5% (95% CI: [97.1, 97.9]) and a false positive<br>rate of 0.8%, which is 1.4% better than Transformer-CAN. Our ablation study demonstrates that removal of<br>the attention mechanism reduces recall by 15%. Statistical significance (p < 0.01) and cross-domain testing<br>in 5G slice data confirm robustness, making it applicable to safety-critical automotive systems.</p>