Evaluating NGBoost as a Model for Probabilistic Prediction for V-Band Power Attenuation

Abstract

A novel application of probabilistic prediction for estimating mm-wave attenuation due to varying weather conditions is developed. Validation of atmospheric propagation models in the W/V-bands has progressed greatly in past years, but the confidence of their predictions has not been validated. The Natural Gradient Boosting (NGBoost) algorithm is tested against a deep neural network to estimate atmospheric attenuation on a 72 GHz terrestrial link and to demonstrate the ability to produce prediction confidence estimates.