Vehicular communications are becoming increasingly
important due to the need for safer, more efficient, and
sustainable transportation. They require the development of
accurate radio channel models for vehicular environments. This
study compares vehicle-to-infrastructure path loss predictions
obtained using four machine learning models: artificial neural
network, support vector regression, random forest, and gradient
tree boosting. The model design employs predictors from the profile
environment between the transmitter and receiver positions.
A methodology to select the best predictor subset is applied by
examining their contribution to performance and interpretability.
We propose a generalization test considering unknown streets
(scenarios), and the results demonstrate that the gradient tree
boosting model significantly improves the path loss prediction
compared to the log-distance path loss model.