Smartphone Application Usage Prediction Using Cellular Network Traffic

摘要

The exponential rise of cellular network traffic demand due to an increased use of hand-held devices requires optimized methods to plan and deliver the necessary network bandwidth. In this paper, we propose to use cellular network traffic generated by smartphone applications to predict which applications the user is likely to be using. We conducted two experiments to assert such feasibility. In one controlled experiment required the users to use applications according to heuristic application usage guidelines. In the other experiment, the subjects were encouraged to use their phone as they normally would. In all cases, we recorded the session time, the uplink and downlink traffic and which applications are running. We subsequently used machine learning algorithms to assess the feasibility of predicting the running applications. We achieved 99% accuracy in the controlled traffic experiment. However, the performance was much lower in the arbitrary traffic monitoring experiment. This preliminary analysis may suggest that it could be possible for cellular network providers to predict what application users are running based on their real-time network usage. This would be in turn used for cellular network optimization and planning.