A Machine Learning-Enhanced Responsive Hybrid Routing Protocol for Throughput Optimization and Link Reliability in Vehicular Ad Hoc Networks

Abstract

The development of vehicular ad hoc networks (VANETs) is central to the creation of intelligent transportation systems (ITSs). VANETs provide the backbone for several applications, some of which are crucial from the viewpoint of safety and necessitate both reliable and high-speed communication. However, given the pronounced characteristics of mobility and dynamic topology in VANETs, link breakage becomes an increasingly common phenomenon, causing packet loss and a decrease in the network throughput. The current paper presents ML-RHR as a new hybrid routing protocol based on machine learning, which improves upon the Responsive Hybrid Routing scheme through the implementation of link-quality analysis. ML-RHR employs the AODV approach to route discovery and relies on the proactive approach to topology maintenance, implemented with MPR as in the Optimized Link State Routing. To enhance ML-RHR performance, link stability is predicted through Random Forest-based models built on SUMO-generated mobility data, which is freely available on the Internet. ML-RHR uses vehicle speed, acceleration, relative distance between two communicating vehicles, and signal strength as features for Link Stability Score and ETX predictions. Simulation experiments, carried out using NS-3 software, allow us to evaluate ML-RHR as compared to AODV, OLSR, and RHR protocols on 5G-VANET and highway mobility dataset, considering different network densities (20, 50, 100, 150, and 200 nodes). The following parameters were used: PDR, Delay, End-to-End, Throughput, Jitter, and Link Failure Rate. The obtained results demonstrate up to 15% PDR improvement over RHR and up to 25% PDR improvement over AODV and an increase in throughput by 20–35%. Link Failure Rate was reduced by 25–40%. The achieved link stability prediction accuracy varied between 85 and 91%.