USER TRAFFIC COLLECTION AND PREDICTION IN CELLULAR NETWORKS: ARCHITECTURE, PLATFORM AND CASE STUDY

With the development of advanced telecommunication technologies and the rapid evolution of mobile communication system standards, radio resource allocation is becoming one of the core issues for the research on mobile communication systems. In this paper, a novel platform architecture for user data collection and traffic prediction is proposed, which consists of the mobile data collection subsystem and the traffic prediction subsystem. The mobile data collection subsystem collects real-time traffic log data of the cellular subscribers from mobile terminals, and the traffic prediction subsystem based on the open source MapReduce framework Hadoop predicts traffic in different time scales by utilizing the data received from mobile terminals. MapReduce framework can improve the computing performance and scalability of the whole architecture. Meanwhile, the support vector regression algorithm (SVR) used in predicting traffic flow can make the traffic prediction more flexible for its remarkable generalization performance. We deploy a platform according to this architecture, and case study shows that this platform can meet the needs of mass traffic processing and achieve high traffic prediction accuracy.