Estimating Quality-of-Service in Urban Vehicular Networks Through Machine Learning

Machine Learning (ML) has emerged as a promising tool for addressing complex challenges in multiple domains. In the context of Vehicular Ad-Hoc Networks (VANETs), ML has gained much more attention due to its ability to solve major known problems in areas such as traffic management, road safety and communication infrastructure management. In a VANET, vehicles generate a significant amount of data, which can be explored to, for example, enhance the network management regarding the connectivity between the vehicles and the infrastructure. This work studies the performance of ML models regarding the estimation of the Quality-of-Service of different network access technologies (ITS-G5 and 5G) in urban vehicular environments. To this end, data collection campaigns were carried out throughout the city of Aveiro, Portugal, which included vehicular and network performance data for ITS-G5 and 5G cellular technologies. After an initial characterization of the data collected, several ML algorithms were trained, considering different combinations of features (represented by the collected metrics). The results have shown that, for the same configurations, similar estimation errors were obtained by the Random Forest Regression and the Extreme Gradient Boosting algorithms, with the last one presenting a shorter estimation time. The results also show that location-independent configurations, i.e., when no geographic positions are used in the ML model, are slightly worse than GPS-based ML models, creating the possibility of being applied in different urban environments, making them quite versatile.