Optimization Algorithms: Who own the Crown in Predicting Multi-Output Key Performance Index of LTE Handover

Abstract

The Long-Term Evolution network (LTE) has been introduced to cater to the rich content applications of multimedia services. With its ability to support lower latency and higher Throughput, the LTE network can provide faster data download speeds. However, once the mobile user moves from one location to another, the performance tends to degrade. Thus, it required the handover from the serving base station to the target base station. Therefore, the telecommunication service providers must provide a further service enhancement to increase the network quality. As a result, the Key Performance Index (KPI) modeling and predictions can be utilized to achieve this objective. In this article, the Extreme Gradient Boosting regressor algorithm has been selected. However, the hyper-parameter associated with this algorithm needs to be optimized first to produce good prediction results. Three optimization algorithms have been chosen: the Annealing Search, Random Search, and the Tree Parzen Estimator. The experiment results show that the Extreme Gradient Boosting with Annealing Search outperformed the Random Search and the Tree Parzen Estimator by producing the lowest MAE and RMSE and higher R2.