A hybrid CNN-NGBoost model for probabilistic image-driven path loss prediction

We propose a hybrid model for probabilistic path loss prediction, based on the footprint of the urban built-up area. A Convolutional Neural Network (CNN) is being deployed in order to extract information regarding the built-up area in the form of a feature vector. The extracted features are then processed from a Natural Gradient Boosting (NGBoost) regressor, who is inherently capable of performing probabilistic prediction. That is, along with point estimations of the received path loss value, the CNN-NGBoost model additionally calculates the prediction interval which covers a user-defined percentage of the prediction distribution. The proposed model can therefore assist network engineers in calculating the risk of their decisions about network coverage.