Due to the growing cybersecurity threats, the privacy and security of modern vehicles have become increasingly significant. Cyber offenses in vehicular networks are recognized using an Intrusion Detection System (IDS). The rapid growth of multimedia technologies has made cybersecurity an essential concern.. Automated and internet-connected cars are exposed to spoofing and jammer attacks. GPS location spoofing is an imminent danger to Connected and Autonomous Vehicles (CAV), threatening security and even exposing motorists and pedestrians to risk. Because of flawed protocol design and increased interconnectivity among modern autonomous vehicles, the Controller Area Network (CAN) bus is insecure. The identification of spoofing attacks on the CAN bus is critical. Hence, it is necessary to develop an efficient spoofing attack detection method to address the limitations of existing models. The major phases of the developed framework are (a) Data Collection, (b) Data Pre-processing, (c) Weighted Feature Selection, and (d) Attack Detection. At first, essential data utilized for the validation is collected in the standard dataset. Further, the gathered time series data is given as input to the data pre-processing phase. Later, the attained pre-processed data is utilized to collect the essential features. Further, in the weighted feature selection phase, the Restricted Boltz-mann Machine (RBM) technique is utilized to attain the significant features. Unlike the conventional CMPA, the proposed FE-CMPA introduces a fitness-entrenchment mechanism that improves the optimization of RBM feature weights and enhances relief score maximization. Subsequently, the acquired weighted RBM features are provided for the attack detection phase. Fur-Furthermore, the spoofing attacks are detected using the developed Hybrid Dilated and Attention-based Network (HD-ANet), which holds the Deep Temporal Convolution Network (DTCN) and Residual Long Short Term Memory (RLSTM) network for effective validation. Hence, the implemented spoofing attack detection model is more secure and achieves a comparatively higher detection rate than traditional approaches in various experimental evaluations.
Deshmukh et al. (Thu,) studied this question.