Hybrid Machine and Deep Learning Framework for Secure Fog-based VANETs

As a conventional implementation of intelligent transportation systems (ITS), a Vehicular Ad-hoc Network (VANET) facilitates smart communication between vehicle nodes and network infrastructures. Nevertheless, the distributed and dynamic characteristics of VANETs make them vulnerable to numerous cyberattacks like DOS and spoofing. Rapid detection of these threats is crucial. It is essential to respond promptly to mitigate threats before they impact vehicle coordination or infrastructure functionality. In this context, fog computing presents an effective solution. Transferring computational tasks from centralized cloud servers to fog nodes allows for quicker detection within the VANET environment. This research utilizes computational resources for swift and localized attack detection. The study leverages the CIC IoMT dataset 2024 to identify various types of attacks at the packet level. Three types of classifiers are utilized. The first classifier assesses whether a packet is benign or malicious. The second classifier identifies which of the 6 specific classes the packet belongs to. The third classifier categorizes the overall type of attack and recognizes the particular protocol employed. Random Forest and Extreme Gradient Boosting are implemented. Additionally, deep learning models are incorporated in this study. The performance of the first classifier ranges from 91.5% to 98% in precision and F1-score, and in recall from 92% to 99%. The results of the second classifier show a precision range of 98.5% to 95%. In terms of recall and F1-score, it ranges from 83% to 95%. The results of the third classifier yield scores ranging from 84% to 99% for precision, recall, and F1-score.