View article

This paper addresses the prediction of Line-of-Sight (LoS) in Manhattan-like urban topologies. A classifier based on the Gradient Boosting Decision Tree (GBDT) model is used for this purpose, whereas training and testing datasets are generated using a standard Ray Tracing algorithm. Compared to a previous work on the same topic, the confidence of the prediction provided by the classifier is compared to an optimized decision threshold, improving the reliability of the prediction with respect to the ground truth, represented by Ray Tracing simulations. Finally, LoS curves as a function of the distance are provided for microand macrocellular cases and different configurations of the environmental features. As a reference, results are compared with the WINNER LoS model. In the microcellular case, it is shown that that the LoS probability curve tends to become more similar to the WINNER one as the height of the base station increases.