Enhancing security in VANETs with unsupervised learning-based real-time false information detection

Vehicular Ad-hoc NETworks (VANETs) enable communication among vehicles and between vehicles and infrastructure to provide safety and comfort to the users. Securing VANETs from malicious nodes is critical, considering their potential impact on public safety. False information detection is one such important security issue in VANETs, where malicious nodes may broadcast bogus information to create the impression of a fake emergency event, like a crash or road congestion. Additionally, multiple malicious nodes may collude to collectively launch false information attacks to increase the credibility of an attack. Detection of false information attacks is critical to mitigate the potential risks they bring to the safety of users and other entities on the road. Existing techniques for detecting false information attacks in VANETs use different approaches such as machine learning, statistical methods, blockchain, trust scores, etc. These techniques rely on historical information about vehicles, past traffic data to train machine learning models used in certain techniques, or coordination among vehicles. Moreover, most of the approaches overlook collusion among attackers. To address these limitations, this dissertation proposes false information attack detection techniques for VANETs using unsupervised machine learning techniques. The objectives of this research are to detect isolated and colluding false information attacks in VANETs with high accuracy and low processing delay using only the real-time characteristics of the network without depending on historical information about vehicles or any past traffic data. The performance evaluation results show that the proposed techniques offer higher accuracy, lower data processing delays, and lower false positive rates compared to existing approaches in scenarios with high proportions of malicious nodes.