Predictive Analysis of Future Cellular Coverage Network Operator, and Vegetation Maps Through Machine Learning : An FCNN Model to predict 6 targets on 6 years of Grid Based Dataset

Sammanfattning

Cellular networks are currently one of the main ways that people connect with each other. In today's digital world, these networks are essential for enabling data exchange and communication. Ensuring the effective growth and optimization of these networks is crucial given the rising demand for high-speed data services. Effective infrastructure planning, improving service quality, and filling in any current network coverage gaps all depend on anticipating future coverage needs (Xue & Liang, 2025). Numerous factors, including the land's topography, the network infrastructure already in place, and even the presence of vegetation in the area, affect how much coverage cellular networks offer (Ruben Borralho, 2021). This study uses geographic, network operator, and vegetation mapping data along with machine learning algorithms to overcome the difficulties in cellular coverage prediction. The goal is to help network operators make well-informed decisions that will lead to better planning strategies and improved service delivery by utilizing these technologies. To accomplish this, a large dataset of ~19 GB of Sweden from 2013 to 2019 was used. Before deciding to train Fully Connected Neural Networks (FCNN) for 6 distinct coverage targets, such as reaching 10 Mbps at 16 dB, a number of machine learning models were tested, including Random Forest, Convolutional Neural Networks and Thresholding techniques. To improve prediction accuracy, preprocessing methods like quantile normalization, Yeo-Johnson scaling, win-sorization, etc were used on the data before the FCNN models were trained. The study's findings showed that the FCNN model achieved an impressive accuracy rate of 93.92% for the initial coverage target. Additionally, there was a significant correlation between the actual coverage seen in 2017 and the coverage maps that were predicted. This accomplished result emphasizes how important it is to take into account elements such as vegetation density and land characteristics when figuring out how cellular signals spread and how to best optimize network coverage for improved performance. This study offers important insights that can help network operators make well-informed decisions about resource allocation and network expansion by developing a robust predictive model that can scale. An important development in network planning is the combination of map data and machine learning methods, which provides a more data-driven strategy for improving the caliber and coverage of cellular networks.