Next-cell Prediction Based on Cell Sequence History and Intra-cell Trajectory

In this work, we study a novel mobility prediction algorithm based on past long-term and short-term trajectories of users. In particular, we perceive the regularity of users’ movements by training a Markov Renewal Process (MRP) using the long-term trajectory history. Moreover, short-term trajectory data, recorded within the current residing cell, is utilized to incorporate possible randomness of users’ behavior into the algorithm. In fact, each neighboring cell is assigned two distinct probabilities of being chosen as next crossing cell, one given by MRP, while another is obtained from the direction of movements across the current cell. Lastly, assigned probabilities, i.e. the pieces of information extracted from the two aforementioned trajectory data sets, are combined with the aid of Dempster-Shafer theory to reach the best possible decision about the future crossing cell. Simulation results illustrate that the algorithm reliably predicts the next crossing cell with around 70% accuracy.