Effects of Data Preprocessing Methods on Addressing Location Uncertainty in Mobile Signaling Data

Recent years have witnessed an increasing use of big data in mobility research. Such efforts have led to many insights on the travel behavior and activity patterns of people. Despite these achievements, the data veracity issue and its impact on the processes of knowledge discovery have seldom been discussed. In this research, we investigate the veracity issue of mobile signaling data (MSD) when they are used to characterize human mobility patterns. We first discuss the location uncertainty issues in MSD that would hinder accurate estimations of human mobility patterns, followed by an examination of two existing methods for addressing these issues (clustering-based method and time window–based method). We then propose a new approach that can overcome some of the limitations of these two methods. By applying all three methods to a large-scale mobile signaling data set, we find that the choice of preprocessing methods could lead to changes in the data characteristics. Such changes, which are nontrivial, will further affect the characterization and interpretation of human mobility patterns. By computing four mobility indicators (number of origin–destination trips, number of activity locations, total stay time, and activity entropy) from the outputs of the three methods, we illustrate their varying impacts on individual mobility estimations relevant to location uncertainty issues. Our analysis results call for more attention to the veracity issue in data-driven mobility research and its implications for replicability and reproducibility of geospatial research.