Abstract
The surge in the number of driverless cars highlights the necessity for enhanced transportation safety and efficiency. Achieving fully autonomous driving depends on the ability of vehicles to comprehend environmental conditions and respond to collective behaviors facilitated by communication technology. Connected and Automated Vehicles (CAV) rely on vehicular ad hoc networks (VANET) to exchange information among vehicles and with infrastructure such as RoadSide Units (RSU). While conventional security measures fall short in the dynamic and uncharted VANET field, machine learning presents a promising avenue for fortification. However, the scarcity of data, particularly regarding Vehicle-to-Infrastructure (V2I) communication, poses a significant challenge for applying machine learning, particularly deep learning. Existing efforts have primarily focused on synthesizing datasets for vehicle-to-vehicle (V2V) communication, leaving a void in the V2I security research. This study bridges this gap with the development of GenVRAM, a simulated CAV dataset generator tailored for creating both normal and attack data, with a specific focus on misbehaving RSU. Through rigorous experimentation with various machine and deep learning models, our research elucidates the inadequacies of traditional machine-learning algorithms in addressing VANET security concerns. Although deep learning exhibits promise, further refinement, and investigation are imperative to ascertain its efficacy in safeguarding VANET. GenVRAM is a pioneering contribution in the realm of VANET security, providing researchers and practitioners with a valuable tool for advancing the resilience of autonomous vehicle systems against malicious attacks.