A Hybrid Machine Learning based Approach for Adaptive Routing in Vehicular Ad-Hoc Networks

Abstract

Vehicular Ad-Hoc Networks (VANETs) are the foundation of communication for the smart transportation of the future, but their efficiency is frequently hindered by the rapid change of topology and high mobility. An intriguing dilemma is that an effective routing process is affected by both fleeting topological transitions and high mobility. To mitigate the issues in this paper, a two step solution is proposed that integrates Machine Learning (ML) and Reinforecement Learning (RL). Initially, a Random Forest classifier selects the most suitable Cluster Heads, thereby establishing a substantial communication group within the network, which in turn stabilizes it. A Q-learning routing agent is positioned within the clustered shell and trained through rewards for making stable and low-latency choices to discover the best data paths. The proposed adaptive method is verified in a realistic simulation environment with the Random Waypoint Mobility Model. The proposed framework demonstrates exceptional performance, achieving a high Packet Delivery Ratio (PDR) averaging 79% and an average end-to-end delay of 0.0004 seconds. These results confirm that the integrated approach provides a robust and highly efficient solution for reliable communication in volatile VANETs.