Artificial Intelligence-driven privacy preservation in the internet of vehicles: a comprehensive systematic literature review

The rapid growth of the Internet of Vehicles (IoV) and the increasing use of Connected and Autonomous Vehicles (CAVs) are causing many privacy and security problems that traditional protection methods can not handle. IoV networks have a lot of different types of data flows, distributed infrastructure, and real-time communication needs. This makes it very hard and important to protect privacy. The constant sharing of data between automobiles, roadside infrastructure, and cloud platforms makes IoV ecosystems vulnerable to attacks that try to manipulate the data, steal it, or make guesses about it. This paper seeks to deliver a systematic and thorough assessment of Artificial Intelligence (AI)-based privacy-preserving strategies in the IoV, tackling the absence of a cohesive framework in existing literature. We employ a Systematic Literature Review (SLR) to examine and classify essential methodologies, such as Federated Learning (FL), Differential Privacy (DP), Homomorphic Encryption (HE), Blockchain-enhanced AI, Adversarial Machine Learning (AML), and anomaly detection, into a six-domain taxonomy. This taxonomy provides a systematic framework for evaluating the advantages, compromises, and scaling constraints of different methodologies. The findings indicate that although AI-driven methodologies have considerable promise in protecting vehicle data, substantial obstacles remain regarding computational overhead, real-time adaptation, and scalability in high-mobility environments. Our work systematically maps existing approaches, revealing performance constraints and unresolved research gaps while delineating specific future research trajectories, including lightweight model design, hybrid privacy frameworks, and integration with emerging technologies such as edge intelligence and 6G networks. The examination of Quality of Service (QoS) and Quality of Experience (QoE) factors indicates that computational overhead (25%), scalability challenges (20%), latency issues (15%), accuracy trade-offs (15%), optimization requirements (15%), and deep learning model complexity (10%) persist as the primary impediments to the real-time, large-scale implementation of privacy-preserving AI in vehicular networks.