Identifying stops from mobile phone location data by introducing uncertain segments

Identifying stops is a primary step in acquiring activity-related information from mobile phone location data to understand the activity patterns of individuals. However, signal jumps in mobile phone location data may create "fake moves," which will generate fake activity patterns of "stops-and-moves." These "fake moves' share similar spatio-temporal features with real short-distance moves, and the stops and moves of trajectories (SMoT), which is the most extensively used stop identification model, often fails to distinguish them when the dataset has coarse temporal resolution. This study proposes the stops, moves, and uncertainties of trajectories (SMUoT) model to address this issue by introducing uncertain segment analysis to distinguish "fake moves" and real short-distance moves. A real mobile phone location dataset collected in Shenzhen, China is used to evaluate the performance of SMUoT. We find that SMUoT improves the performance (i.e., 15 and 19% increase in accuracy and recall rate for a one-hour temporal resolution dataset, respectively) of stop identification and exhibits high robustness to parameter settings. With a better reliability of "stops-and-moves" pattern identification, the proposed SMUoT can benefit various individual activity-related research based on mobile phone location data for many fields, such as urban planning, traffic analysis, and emergency management.