A New Approach for Line of Sight Prediction with Geometry Analysis and Machine Learning in Diverse Environments

The purpose of this paper is to provide a highly accurate and efficient framework that enables the prediction of Line Of Sight (LOS) and Non-Line Of Sight (NLOS) probability between a transmitter
$(T_{x})$
and a receiver
$(R_{x})$
. We propose the implementation of a hybrid system that consists of geometry based model such that the environment is thoroughly described with the purpose of extracting the essential geometrical characteristics of the environment, where the main components of the channel modeling such as LOS are affected by these character-istics within the radio propagation scenario, these attributes are transformed to features that are pre-processed along with labels (true values) generated from ray tracing for each environment. This dataset is used as input to a Machine Learning (ML) based Gradient Boosting Decision Tree (GBDT) binary classifier, that outputs a prediction for either LOS or NLOS. The proposed model achieves acceptable results on multiple environments with fraction of the time required by ray tracing.