<p>The development of the Internet of Vehicles has created overwhelming challenges regarding communication efficiency, data privacy, and resource allocation. Existing federated learning models have high communication overhead (more than 40%) and privacy threats because of the inefficient formation of clusters. To solve these challenges, this proposed research undergoes three stages, which are cluster formation, cluster head selection, attack detection, and security. Initially, the Clustered Federated Learning architecture is utilized to improve communication effectiveness by grouping vehicles into clusters by feature similarity. This structured methodology avoids redundancy and guarantees maximized data merging over traditional FL, which is plagued by excessive latency in decentralized settings. Secondly, the cluster head is selected with the aid of a Hybrid Fungal Growth Chinese Pangolin Optimizer. Thirdly, the Enhanced Secure On-Demand Distance Vector algorithm is used to detect malicious nodes, prevent message manipulation attack, eavesdropping attack, message holding attack, and black hole attack, then provide secure transmission in the Internet of Vehicles. Propose a Context- and Mobility-Aware Forwarding strategy using a Deep Recurrent Neural Network algorithm to maintain an up-to-date Neighbor Table. This approach efficiently predicts and leverages node trajectories for optimal packet forwarding. Finally, a human–machine interface is introduced for sending warning messages to the drivers for improving road safety by monitoring various road behaviors and traffic conditions. The proposed system is simulated in the network simulator version 3 platform, and the experimental outcome illustrates a throughput of 99.23% which is excellent when compared with other traditional algorithms.</p>