Sybil Attack Detection in VANETs: An LSTM-Based BiGAN Approach

The growing prominence and evolution of the Internet of Vehicles demand more prompt and precise security measures. Real-time detection of Sybil attacks in Vehicular Ad-Hoc Networks (VANETs) has emerged as a critical challenge. We propose an intrusion detection scheme for VANETs based on Long Short-Term Memory Bidirectional Generative Adversarial Networks (LSTM-based BiGAN) to tackle this issue. Our method employs LSTM to process time series in Basic Safety Message (BSM) information and trains normal data using BiGAN. Considering the diversity of BSM data, we introduce a multivariate Gaussian distribution as input to better align with the actual data distribution. Furthermore, we develop an Abnormal Score (ANS) algorithm, which assesses whether a BSM originates from a normal or abnormal vehicle. Our research successfully detects five attacks in the Veremi dataset, achieving an impressive average accuracy rate of 93%.