Securing Automotive Networks from DoS and Fuzzy Attacks with Optimized LSTM Models

Intelligently connected automobiles have come a long way thanks to the deep integration of cutting-edge networked gadgets and advancements in automotive technology. The increasing connectivity of vehicles has introduced significant cybersecurity concerns, particularly in controller area network (CAN) bus systems, which are crucial for vehicle communication. For this, the research presents a novel hybrid model combining and integrating a long short-term memory (LSTM) neural network and bacterial foraging optimization (BFO) technique to address challenges like identifying fuzzy and denial of service (DoS) attacks on the CAN bus system. The model’s efficacy is demonstrated by evaluating various cyber-attacks from the Car-Hacking dataset. An adaptive feature selection method using BFO to identify optimal CAN bus characteristics for accurate attack detection. The LSTM-based temporal pattern recognition system detects anomalous message sequences and real-time countermeasures for mitigating DoS and fuzzy attacks. Reducing attack detection time to 0.0838 s, an enhancement over LSTM-AE, suggests improving detection speed. With a higher precision of 94.6% and F1-scores of 95.8%, LSTM-BFO outperforms other models based on accuracy, precision, and F1-score under the current experimental setup.