GAUSSIAN PROCESS WITH HYPERPARAMETERS OPTIMIZATION USING LBFGS ALGORITHM: A CASE STUDY WITH 5G NEW RADIO THROUGHPUT DATA

Gaussian Process Regression (GPR) is a fast and powerful non-parametric regression method for data mining and machine learning. The Bayesian optimization method, which has remained one of the standard methods of optimizing the GPR, usually leads to poor parameter tuning and code start problems. In this paper, we proposed and leveraged the accurate and robust gradient-based limited-memory Broyden–Fletcher–Goldfarb–Shanno (LBFGS) algorithm to surmount the aforementioned Bayesian optimization tuning method. We have applied the proposed GPR-LBFS tuning algorithm to mine and predict a set of throughput data that were acquired over 5G New radio networks. We show by engaging the Root Mean Square Error (RMSE) and Correlation coefficient (R) statistics, that the proposed GPR-LBFS tuning algorithm provides the best hyperparameter tuning results, and also attains the best throughput data prediction accuracies at different measurements points and spatial domains