Machine learning for LTE network traffic prediction

Predicting network traffic is key to resource management and next-generation network planning. Conventional traffic prediction was based on statistical autoregressive models, which found correlations between current and projected traffic levels. Such models have severe limitations, however; specifically, most of them only work on stationary statistics, which limits them to basic networks and slowly changing levels of incoming traffic. Most of the state-of-the-art networks, however, have a sophisticated structure and rapidly fluctuating traffic flows. Energetic efforts have recently been made to bring machine learning to bear on the limitations of conventional models and to improve predictions. The paper examines key machine learning solutions based on artificial neural networks versus conventional statistical approaches, as well as their practical applications for LTE network traffic prediction.