What Machine Learning Predictor Performs Best for Mobility Prediction in Cellular Networks?

It is envisaged that the future cellular networks (5G) will be able to meet the promising capacity and quality of experience requirements through extreme network densification and conglomeration of diverse technologies. It is easy to fathom that efficient management of such a convoluted network will be one of the big challenges faced by 5G. To cope with this challenge, Self-Organizing Networks (SONs) that were originally designed for legacy networks with reactive approach needs to be transformed to proactive paradigm. This radical transformation is possible only if the future network state can be predicted beforehand by harnessing historical network data. Mobility prediction is one of the key enablers of Proactive SON which enables efficient resource management. In this paper, we perform comparative analysis of four mobility predictors: Deep Neural Network (DNN), Extreme Gradient Boosting Trees (XGBoost), Semi-Markov, and Support Vector Machine (SVM). Our investigation is based on realistic synthetic dataset of eighty-four mobile users generated through realistic Self-similar Least Action Walk (SLAW) mobility model. We evaluate the effectiveness of each model not only based on the model's ability to predict the future location of mobile users but also the time each algorithm takes to be fully trained and perform such prediction. XGBoost stands out as clear winner among all predictors considered with high accuracy of 90%. Its high prediction accuracy enables high energy saving gain of above 80% when it is employed for driving proactive energy saving SON solution.