QoS Prediction in V2X Networks Using Decision Tree Ensemble Methods Under Different Spatial Training Strategies

QoS prediction can improve reliability, adaptability and decision-making in V2X networks. In this paper five ensemble machine learning methods based on Decision Tree regressors were evaluated, analyzing their performance when applied to a QoS prediction problem, under three different spatial training split strategies, obtaining training and testing time, MAE and R² scores. Training and testing were performed using a public dataset, with measurements collected from vehicles across different urban areas. All evaluated models are available in the Scikit-Learn platform. Results indicate that Histogram-Based Gradient Boosting (HGB) provides the best trade-off between computational cost and prediction accuracy for large V2X datasets, while balanced splitting strategies improve generalization when representative training samples are preserved.