SHORT-TERM MOBILE DATA TRAFFIC FORECASTING: A CASE STUDY OF KADUNA STATE, NIGERIA

Mobile networks are essential for today's seamless communication. As more individuals subscribe to mobile networks, the need for mobile networks has increased significantly. The network operators must devise strategies to handle the enormous demand of mobile network resources, such as spectrum, which are costly. There is a need for effective network resource management as well as a mechanism to predict future networks that can be used for network management and planning. This study uses real-life data to forecast mobile traffic using Kaduna State as a case study and compared prediction algorithms with the hybrid. The data set was gotten from a network provider. The hybrid combination of LSTM and AGB has been proposed in this study, and its performance has been compared with LSTM and ARIMA using MAE, RMSE and MAPE as evaluation metrics. The prediction performance of the algorithms was carried out on ten base stations with both the highest and lowest traffic from two local government areas, which are Kaduna South and Kaduna North. The LSTM-AGB outperformed LSTM and ARIMA. From the performance evaluation, the RMSE, MAPE and MAE of all the selected base stations in LSTM-AGB have a lower value than LSTM and ARIMA, which indicates a good fit of the model. It was observed that the hybrid algorithm performed better in base stations with high traffic