Anomaly Detection in IoV Can Bus Traffic Using Variational Autoencoder-LSTM with Attention Mechanism

The rapid growth of the Internet of Vehicles (IoV) greatly enhances the communication and data exchange capability between vehicles, which allows the real-time transmission of critical information. The CAN bus is the backbone developed to transmit data between Electronic Control Units in the in-vehicle networks and it is prone to various cyberattacks that can breach its security and integrity. To identify cyberattacks on a CAN bus this paper proposes a deep learning model employing a Variational Autoencoder in conjunction with Long Short-Term Memory (LSTM) that features an integrated attention mechanism. The proposed model leverages the temporal characteristics of CAN bus traffic to detect anomalies by identifying the differences in the reconstruction loss. The CICIoV2024 dataset, containing benign and malicious messages for the CAN bus, was used for the proposed model, which achieved high effectiveness with the accuracy of 99.78%, precision of 99.81%, a recall of 99.32%, and F1 score of 99.56%. The results show that the model is strong in accurately detecting unusual behaviour and offers a flexible solution for improving cybersecurity in IoV systems.