In modern telecommunication networks, traffic congestion and overload on cells pose a serious challenge to maintaining service quality. While nominal capacity values for each cell are often known, operators typically lack accurate insights into how much traffic exceeds these thresholds in congested areas. In this study, an approach is proposed to estimate the excess traffic load on each cell using trajectory data. The core idea is that dynamic population density, derived from the spatial distribution of users, plays a crucial role in shaping traffic patterns. To approximate each cell’s coverage area, circular approximation is employed, allowing users to be associated with their nearest cell based on spatial proximity. By combining user presence per cell and hour with simulated traffic values, a comprehensive dataset is constructed and fed into a Random Forest Regressor model. The trained model is used to predict actual traffic, and overload conditions are detected by comparing predicted values with nominal capacities. The proposed framework enables identification of congestion-prone cells and supports informed decision-making for network infrastructure expansion.