Multi-environment automotive QOS prediction using Machine Learning for Vehicle-to-Anything (V2X) communications

The advent of 5th Generation (5G) wireless systems has ushered in a new era of vehicular applications, where reliability and low latency are paramount. This evolution is expected to continue with the transition to 6G and beyond, heavily relying on Machine Learning (ML) to make proactive decisions and manage the Quality of Service (QoS) for an enhanced user experience. Particularly in vehicular communications, the Vehicle-to-Everything (V2X) schemes stand to gain immensely from these technological advancements. In this study, we explore this cutting-edge domain and propose LightGBM-enhanced MLP (LeMLP), a lightweight and effective ML approach focusing on the prediction of QoS in V2X communications using LightGBM and Multilayer Perceptron (MLP), a critical component for the safety and efficiency of autonomous vehicles. Our methodology builds upon the Berlin V2X dataset, encompassing diverse urban environments' data sources, including GPS, ping, iperf, and LTE. In our proposed technique, the model is trained on data from one mobile network operator and then evaluated on data from another (multi-domain approach), reflecting real-world scenarios of varying network conditions. Through this multi-domain adaptation approach, LightGBM's split method, and fine tuning, we achieved an R2 score of 0.9447 on evaluation using only six key features. This improvement is significant in the context of 5G communication systems, where the prediction and early notification of QoS changes can profoundly impact the operational safety of autonomous vehicles. This research not only contributes to the technical advancement in vehicular networks but also opens avenues for exploring various data, feature selection strategies, and prediction qualities. It lays a solid foundation for deploying robust ML models in real-world scenarios, thereby advancing the capabilities of V2X communications in the era of 5G and beyond.