Machine learning based throughput prediction in 5G NR non-terrestrial networks using system-level simulator

The ability to accurately predict a network’s throughput can enable mobile operators to allocate resources efficiently and improve the overall user experience. Nonetheless, the multidimensional parameter space involving such a network-level performance metric requires careful analyses of the underlying parameters and cross-layer dependencies. This paper presents a study on evaluation of classical machine learning algorithms for predicting the throughput in 5G NR Non-Terrestrial Networks using data from system level simulations. The study was conducted in three stages, namely dataset generation, dataset validation and visualization, and model evaluation. The results provide confidence in usage of system simulations for training and evaluation of machine learning algorithms. Furthermore, analyses show that techniques such as Random Forest Regression can prove to be quite effective and robust in throughput prediction.