Optimized Performance Analysis of VANET Routing Strategies Using Machine Learning

Abstract: Vehicular Ad-hoc Networks (VANETs) have the capability to enable smooth Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication, making them a prominent research area in the development of intelligent transportation systems. Nevertheless, rapid and unanticipated changes in network architecture, caused by the highly dynamic and varied mobility patterns of vehicles in densely populated metropolitan environments, present major challenges. These challenges not only hamper the effectiveness and reliability of routing protocols under such unstable conditions but also affect consistent and dependable data transfer between vehicle nodes. This article introduces machine learning to identify the optimal routing protocols considering constraints such as vehicle speed and varying population density. The method selects the best routing protocol based on its performance, aiming for improved outcomes in terms of throughput, Packet Delivery Ratio (PDR), and End-to-End (E2E) delay. The study models an actual urban setting using OpenStreetMap (OSM), while the Simulation of Urban Mobility (SUMO) framework is employed to simulate traffic dynamics. Among all evaluated models, Gradient Boosting combined with Ad-hoc On-Demand Distance Vector (AODV) consistently outperformed others in terms of accuracy (up to 100%), PDR (94-96%), and minimal E2E delay. In contrast, Support Vector Machine (SVM) with AODV achieved peak throughput under lower vehicle densities, demonstrating its strength in light traffic conditions.