Site-Specific Machine Learning Approach for Line of Sight Detection

The purpose of this paper is to analyze the problem of predicting whether or not a transmitter and a receiver are in the Line-Of-Sight (LOS) condition. Different scenarios are considered according to different city maps and the transmitters' and receivers' locations. This paper proposes an algorithm for the prediction of LOS that utilizes Gradient Boosting Decision Trees (GBDT) as a baseline classifier. It is a binary classification machine learning algorithm. For training and testing the GBDT classifier, a synthetic ray-tracing dataset of the real, complex environment is generated. The generated datasets from RT can be used to train, validate and test the model. According to the results, the GBDT model provides accurate estimates of LOS probability and achieves effective classification performance for unseen data to the model during the training. It can be inferred from the estimation of the importance of the features that the model learned simple decision rules that conform with the physics behind the evaluation and are intuitive.