Hybrid Optimized Noise-Suppressing Zeroing Neural Network-based Cyber-Attack Detection in Autonomous Vehicles

Autonomous vehicles (AVs) are on the horizon and are expected to significantly improve transportation safety and comfort. These vehicles are linked with different external systems and use advanced embedded technologies to notice their surroundings and make informed decisions. Traditional cybersecurity mechanisms are often insufficient for AVs due to their unique real-time processing requirements, dynamic environments and heterogeneous system architectures. To overcome these complications, Hybrid Optimized Noise-Suppressing Zeroing Neural Network-based Cyber Attack Detection in Autonomous Vehicles (CAV-NSZNN-HSDOA) is proposed. The input data are gathered from the load testing database. The gathered data are supplied to the preprocessing stage. During preprocessing, Region-Aware Neural Graph Collaborative filtering (RANGCF) is used for cleaning the input data. The preprocessed data are fed to the feature selection stage. By utilizing Portia Spider Optimization Algorithm (PSOA), it selects features like Total Length of Forward Packets, Flow Duration, Forward Packet Length Mean and Active Mean. Then, the selected features are transferred into the Noise-Suppressing Zeroing Neural Network (NSZNN) to detect cyberattacks in AVs as normal and anomaly. Finally, Hybrid Sine Cosine and Dipper-Throated Optimization Algorithms (HSDOA) is employed to improve the weight parameters of NSZNN. The CAV-NSZNN-HSDOA technique is executed in Python. The metrics, such as accuracy, precision, recall, specificity, F1-score and ROC, computational time, are examined. The experimental results show that the CAV-NSZNN-HSDOA method outperforms by achieving a higher accuracy of 99.41% and a higher sensitivity of 98.24% compared with the existing techniques.