Optimization of Geographical Fingerprinting for 5G Networks Using Machine Learning

This thesis enhances network optimization for 5G networks through a Machine Learning (ML)-driven approach to geographical fingerprinting. In collaboration with WiLab and a globally renowned telecommunications company, the study utilizes Minimization of Drive Tests (MDT) to collect vital data from User Equipment (UE), crucial for improving positioning accuracy in network sectors with diverse density and mobility patterns, from urban to rural environments. A key advancement is the refined application of the weighted k-Nearest Neighbors (WKNN) algorithm, specifically adapted to enhance localization in varied settings, optimizing performance across the spectrum of network conditions. A significant innovation of this work is the use of synthetic data generated through generative AI (Gen-AI) models to address the limitations of sparse data in challenging environments. By creating high-fidelity synthetic datasets, the study simulates realistic network scenarios, enhancing model training without extensive ground-truth data. This approach significantly improves positioning accuracy within 5G networks, particularly in sparser areas, and reduces reliance on traditional data collection methods. This thesis contributes to both theoretical advancements in network management and practical applications in deploying ML to enhance next-generation wireless networks. The methodologies developed are poised to influence future advancements in 5G technologies, paving the way for more robust and efficient network services