@article {10.3844/jcssp.2026.2963.2974,
article_type = {journal},
title = {An Efficient Cluster Head Selection in IoV-Enabled VANETs Using a Federated Diffusion Convolutional Recurrent Neural Networks Framework},
author = {Subhash, P and Srinivas, Madana and Ranjith, A. and Baireddy, Ravinder Reddy and Qayyum, Mohammed},
volume = {22},
number = {10},
year = {2026},
month = {Oct},
pages = {2963-2974},
doi = {10.3844/jcssp.2026.2963.2974},
url = {https://thescipub.com/abstract/jcssp.2026.2963.2974},
abstract = {The IoV enables the formation of VANETs that require a strong selection of CHs for effective communication within high-velocity changing traffic scenarios. Existing approaches often face issues related to scalability, privacy, and adaptability. To meet the aforementioned urgent needs, we design a decentralized scheme for intelligent CH selection based on Federated Learning with Diffusion Convolutional Recurrent Neural Networks (DCRNNs). According to this proposed methodology, vehicles are responsible for local collection and preprocessing of spatiotemporal mobility data. Next, they proceed to train lightweight versions of DCRNN and exchange only the resultant model updates for global aggregation to ensure data privacy. Then, a distributed voting-based mechanism has been utilized for the selection of optimal CHs, according to predicted connectivity and stability metrics. The proposed work is evaluated in depth through simulations based on NS-3 and SUMO under urban and highway mobility scenarios. The evaluation demonstrates a 40% reduction in network communication overhead relative to centralized alternatives while maintaining high accuracy for CH selection and low decision-making latency. In addition, the framework proves to be scalable even in dense vehicular environments, as we can dynamically reassess CHs based on changes in traffic conditions. The key novelty of the projected research is highlighted with the unified integration approach of federated learning combined with Diffusion Convolutional Recurrent Neural Networks (DCRNNs) to guarantee a framework with prime features like privacy preserving, spatiotemporal awareness, and adaptability in nature for IoV-enabled VANETs to the Cluster Head (CH) selection process. Therefore, the approach promises significant enhancements in its efficiency and robustness. Collectively, this architecture provides a powerful and efficient method for the dynamic communication networks of IoV applications in VANETs.},
journal = {Journal of Computer Science},
publisher = {Science Publications}
}