HyperLSTM for Anomaly Detection Model for VANET Security Using the VeReMi Extension Dataset

Intelligent transportation systems safety and reliability rely on anomaly detection in vehicular ad hoc networks (VANETs). Robust detection methods with learning and response capabilities to different vehicle behaviors and potential security risks are required for the dynamic and complex nature of VANET communications. This research brings methodological innovation into the area through a HyperLSTM model, an extension of the standard LSTM network optimized to deal with increased complexity and flexibility, specifically engineered to cope with the complexity of VANET data. Standard LSTM models are usually poor in high-dimensional time-series data where the HyperLSTM structure is specifically engineered for detecting temporal relationships and anomalies. Two of the primary methodological contributions are a dropout technique for increasing generalizability and the application of overlapping windows for streams of real-time data. The proposed HyperLSTM model surpasses the existing methodology of machine learning and deep learning, which often struggles with accuracy varying from 80% to 95% with an accuracy of 98%. With the error metrics of 0.0649 mean squared error (MSE), 0.2547 root mean squared error (RMSE), and 0.0649 mean absolute error (MAE), the HyperLSTM model achieved significant performance values. This research introduces a dynamically adaptive HyperLSTM framework that extends conventional LSTM capabilities, specifically designed to capture the complex spatial-temporal patterns inherent in VANET communications, which have been less explored in previous studies. This study demonstrates how efficiently HyperLSTM networks leverage VANET anomaly detection, thus enhancing methodological efficiency in managing the complexity and diversity of vehicle network data. Of great significance for future generations of vehicle communication systems, the findings prompt the incorporation of HyperLSTM models into future VANET security systems, thus enhancing detection efficiency.