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

The rapid development of intelligently connected automobiles has been greatly facilitated by the profound integration of technologically advanced networked devices and automotive innovation. The Controller Area Network (CAN) communicates to all connected devices called nodes inside a vehicle without verifying the addresses of those nodes to facilitate network connectivity within the vehicle. Because of this, it is highly susceptible since it lacks the authentication and authorization protocols, leading to all possible attack varieties. By integrating the Bacterial Foraging Optimization (BFO) technique with a Long Short-Term Memory (LSTM) neural network, the system presents a novel idea to identify and avert fuzzy and Denial of Service (DoS) attacks on the CAN bus system. The effectiveness of the suggested method is proven by extensive experiments conducted on real-world car hacking datasets. The related CAN data features optimized for feature selection can be retrieved using the BFO algorithm. The LSTM model's features identify irregular CAN bus activity by learning complicated temporal patterns and capturing long-term dependencies. The methods, like rate limits, filtering IDs of high-priority messages, and preventing malicious congestion, are implemented to avoid DoS attacks. System responses to fuzzy attacks include correcting abnormal communication patterns, avoiding random IDs, and filtering attacked data. The results show that the methodology improves the security of automotive networks by detecting CAN bus attacks with high accuracy, low False Positive Rates (FPR), and efficient attack prevention capabilities.