Securing the Road Ahead: A Comprehensive Review of Hybrid Deep Learning and Optimization-based Intrusion Detection in Vehicular Networks

With the rapid evolution of intelligent transportation systems, the integration of Vehicular Ad Hoc Networks (VANETs), Internet of Vehicles (IoV), Industrial IoT (IIoT), and Unmanned Aerial Vehicles (UAVs) has introduced a new range of cybersecurity vulnerabilities. These networks are increasingly targeted by advanced threats such as Denial-of-Service (DoS), Sybil, spoofing, replay, and distributed attacks, which can disrupt real-time communication, compromise safety-critical functions, and breach driver privacy. To address these challenges, researchers have turned to advanced Intrusion Detection Systems (IDS) powered by hybrid DL and ML techniques. This review systematically evaluates recent developments in IDS for vehicular and IoT networks by analysing multiple cutting-edge studies. The models examined include hybrid CNN-LSTM, GRU with attention mechanisms, transfer learning-based IDS, federated learning architectures, and blockchain-enabled detection schemes to name a few. It was found that Several models achieved exceptionally high detection rates—often exceeding 97% to 99.9% accuracy—for both known and zero-day attacks. Feature selection and optimization techniques like PCA, mutual information, particle swarm optimization, and genetic algorithms were commonly used to reduce model complexity and improve performance. The review identifies several research gaps, including a lack of lightweight IDS deployment strategies for resource-constrained environments, limited generalizability across real-world vehicular topologies, and insufficient adaptability to evolving attack vectors. The paper concludes with recommendations for future research, such as incorporating distributed IDS architectures to enhance the robustness and applicability of IDS in next-generation vehicular networks.