Machine Learning-Based Prediction of Optimal Neighbour Cells for LTE Handover in Dense Urban Areas: A Case Study from San Francisco

Abstract

The demand for uninterrupted and high-speed mobile data continues to grow, driven by the rapid expansion of IoT systems, communication applications, social media platforms, and the increasing number of mobile users. The handover mechanism plays a critical role in maintaining uninterrupted service during user mobility, and indirectly affects data throughput by influencing the device's Reference Signal Received Power (RSRP) levels. In LTE systems—which remain widely used globally—the handover process is vital for ensuring service quality, and its significance is expected to increase further with the densification of base stations in upcoming 5G and 6G technologies.  In this study, we utilize a dense LTE drive test dataset to first estimate the device’s distance to the base station and the geographical locations of base stations. These estimates, combined with parameters such as serving and neighbour cell identities and DL EARFCNs from seven different cells, are then used to develop an efficient machine learning–based handover prediction model. To evaluate and compare the performance of the Random Forest and XGBoost algorithms, multi-class classification metrics including precision, recall, and F1-score were utilized. The results demonstrate that Random Forest model can effectively identify the optimal target cell without the need for traditional, complex handover algorithms. The XGBoost algorithm gave much lower handover performance rates and F1-score compared to Random Forest.