Prediction and analysis of mobile subscribers network traffic using time series algorithms in machine learning

Given the rapid development of information and communication technology, the popularity of mobile internet communications has reached an unprecedented level. Accurate forecasting, understanding, and analyzing network traffic are vital for network planning, performance optimization, fault detection, and security management. With the scaling of networks and the introduction of new services, network traffic exhibits diverse characteristics, including variations, trends, and randomness. These characteristics make accurate traffic prediction a challenging task. In this paper, the Long Short-Term Memory (LSTM) network model is chosen as the proposed algorithm for predicting mobile network traffic. This algorithm is based on Recurrent Neural Networks (RNNs), which are applicable to time series problems, thus commonly used for predicting mobile traffic and capturing temporal dependencies. Finally, after examining various time series algorithms in machine learning, the LSTM model was implemented on a real dataset of approximately ۲۵,۰۰۰ users of the mobile network traffic of the Rightel operator, the third operator in Iran. By optimizing the proposed model, the metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination showed significant improvement compared to other models. These results indicate that the developed model can effectively be used to improve internet management and customer services.  To improve the model’s accuracy, external factors such as promotions and operator updates should be considered during the data labeling process, and we evaluate and improve network traffic using MAE and RMSE metrics.

کلیدواژه ها:

Mobile network traffic prediction ، time series algorithms ، Machine Learning ، Deep Learning ، Long short-term memory network