5G Long-Term and Large-Scale Mobile Traffic Forecasting

A number of factors, such as the ongoing development of more intelligent mobile phones, the introduction of machine-to-machine connections, and the availability of enticing and data-intensive applications, are driving up the demand for mobile data traffic globally. Effective and precise mobile traffic forecasting is particularly important for 5G networks, which are expected to have much higher levels of traffic compared to previous generations of mobile networks.It has been well known that implementing traffic prediction can improve energy efficiency, ease resource allocation, provide the best user experience, and finally enable intelligent cellular networks. Traffic prediction has emerged as one of the main enabling technologies for autonomous networks, which is supported by the whole telecommunication sector, with the large-scale commercial deployment of the 5G network. Additionally, traffic forecasting is a crucial component of numerous transportation services, including navigation, route planning, and traffic control. By dynamically allocating network resources in accordance with actual traffic demand, precise short-term prediction of future traffic load information improves network energy efficiency, while long-term forecasting is crucial for network planning and base station localization. For many practical applications, such as predicting the demand for mobile data traffic, time-series prediction techniques are essential. Generally speaking, there are two types of data prediction models: traditional and machine learning models[1]. Traditional techniques include statistical methods such as Autoregressive Integrated Moving Average (ARIMA) and its extensions, such as Seasonal ARIMA (SARIMA). Due to numerous aspects, such as user mobility, the arrival pattern, and distinct user requirements, the pattern of network traffic is actually highly complex.