A Machine Learning Approach to Forecasting TCCH CongestiA Machine Learning Approach to Forecasting TCCH Congestion in Nigerian GSM Networks: Comparing Supporton in Nigerian GSM Networks: Comparing Support Vector Regression and Gradient Boosting Regression

The rapid expansion of mobile telecommunications in Nigeria has intensified the demand on network infrastructure, resulting in persistent Traffic Channel (TCCH) congestion that undermines service quality and user experience. Accurate forecasting of congestion trends is critical for effective resource allocation, regulatory compliance, and network planning. This study applied two machine learning techniques, Support Vector Regression (SVR) and Gradient Boosting Regression (GBR), to forecast monthly TCCH congestion rates across Nigeria’s four major GSM networks: MTN, Airtel, Globacom, and 9mobile. A nine year dataset spanning January 2015 to December 2023 was obtained from the Nigerian Communications Commission (NCC) and restructured into a supervised learning format using twelve lagged features and month of year dummy variables to capture temporal dependencies and seasonal patterns. Both models were trained on data from January 2016 to December 2023 and employed a recursive multi step forecasting approach to predict congestion for January to December 2024. Model performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results showed that while both models effectively captured monthly variation trends, SVR consistently produced lower error values and aligned more closely with actual observations, particularly for Airtel, Glo, and 9mobile networks. GBR tended to overestimate congestion, especially at low actual values, leading to higher percentage errors. The findings demonstrate the superior accuracy and stability of SVR for short term TCCH forecasting in Nigerian GSM networks. Incorporating such machine learning models into operational workflows can enhance proactive congestion management, optimise frequency resource planning, and support compliance with regulatory benchmarks.