Trajectory pattern change analysis in campus WiFi networks

Mobile devices equipped with geo-location modules facilitate researches on the nature of human mobility. While valuable contributions have been made to discover trajectory patterns from movement history, there are still not a comparative amount of works specialized in analyzing how trajectory patterns change over time. In this paper, we facilitate this problem in a systematic and quantified manner as identifying potential change points of user trajectory patterns extracted from successive time intervals. Specifically, we present a unified, information-based measure to quantify pattern changes between two intervals, and perform a Bayesian analysis on a sequence of aggregate measures for each individual interval to detect the actual change points. Experimenting on a three-month long dataset in real campus WiFi networks, we show that our approach is effective to identify trajectory pattern changes in practice with a discussion on the impact of internal parameters of proposed model on detection performance. Furthermore, we also inspect external factors influencing user mobility in reality by associating trajectory pattern changes with public events, and there shows an interesting connection between group pattern changes and real-world principles such as the weekday calendar or public events such as the May Day holiday.