Predicting Mobility Management Demands of Cellular Networks based on User Behavior

Mobile network operators continuously !aim to optimize network resources to satisfy user demands and be energy efficient. In the era of network function virtualization, predicting the demands for certain functional resources becomes a necessity. Resource optimization is usually based on modeling the user behavior and demands – which gets more challenging if the behavior and demand change dynamically. This applies to the mobile network control plane, where session and mobility management load could have certain seasonality, although the ever-changing user behavior modulates the models. As user behavior keeps varying, traditional mathematical time series analysis cannot provide sub-optimal results. This paper proposes using machine learning methods to predict certain mobility management efforts, such as handovers and tracking area updates. In the cellular network, HO (Handover) and TAU (Tracking Area Update) messages are used for user mobility management, among others. These appear in the S1AP protocol for LTE and 5G NSA (Non-Standalone Architecture). In this paper, we use real-life HO and TAU message records and aim to predict their volumetric appearance for the upcoming time window. We compare the prediction results for a baseline, a linear, a CNN (Convolutional Neural Network), and an LSTM (Long Short-Term Memory NN) model. The ultimate aim of these predictions is to estimate the expected network demands for certain network functions.