GraTree: A gradient boosting decision tree based multimetric routing protocol for vehicular ad hoc networks

Abstract

The Intelligent Transport System (ITS) is increasingly becoming a reality thanks to the incorporation of modern vehicles that, together with artificial intelligence techniques, can bring novel solutions to the smart city. Today, vehicles are embedded with different types of wireless devices that allow them to communicate with a wide range of telecommunication infrastructures located on streets, avenues, or roads. In this context, valuable data related to traffic management is transmitted and can be used for different purposes, such as assisting other vehicles on the road or improving the decisions made by traffic engineers. Vehicular Ad hoc NETworks (VANETs) enable wireless communications between vehicles, between vehicles and infrastructure, and also vehicles-to-everything, with the aim of generating smart mobility solutions and contributing to improve city services. However, routing information in VANETs is a challenging task, as vehicles are highly mobile, and their trajectories are uncertain and highly constrained by the city’s road map. In this regard, as the application of machine learning (ML) algorithms in wireless networks is increasing, the performance of vehicular networks can be improved by applying prediction-based techniques. In this work, we propose a multimetric ML-based routing protocol to select the best next-hop node for forwarding warning messages. For this purpose, we have considered the CatBoost framework, a gradient boosting algorithm on decision trees. Furthermore, we studied the importance of each routing metric and selected only the most relevant ones in our routing decisions. The performance evaluation shows the significant improvements obtained from our ML-based approach in terms of packet losses.