Hotspot analysis and prediction for mobile networking applications

Understanding the dynamics of mobile traffic is highly valuable for a variety of fields, such as transportation and networking. In particular, analyzing hotspots, i.e., areas presenting an increased popularity at certain times, is crucial for adequate planning and management operations. Yet, despite its importance, we lack today a precise definition of the term hotspot in the community. The essence of this contribution is based on a unique mobile phone dataset collected by a French mobile operator in the city of Paris. In this work, we propose a new definition for the hotspot concept while highlighting the major weaknesses of the literature. Moreover, we provide an extensive benchmarking for the hotspot forecasting problem. Our results show that Long Short-Term Memory (LSTM) gives the best performance for the hotspot prediction problem, and we consider it for a Robotic Aerial Base Station (RABS) deployment application. In order to minimize the RABSs’ travel distances, we mathematically model the problem and introduce a greedy and Particle Swarm Optimization (PSO) algorithms to solve it. The results in terms of coverage ratio and travel distance showcase the difference between a prediction-based approach and a non-prediction-based approach.