Exploring 2-D temporal patterns of individual mobility using non-negative matrix factorization

Abstract

Understanding human mobility patterns is important for sustainable mobility development. One practical issue is about observing individual mobility regularity and variability over time and space. After the extraction of pre-treated travel attributes from a geolocation dataset, a further exploration of individual mobility profiles with multiple features is always required. The big challenge on this is to deal with a large number of mobility variables in order to find fine-grained temporal patterns. This paper aims to identify temporal mobility patterns with the two-dimensional (i.e., 2-D) trip attributes, i.e., departure time and distance. We use the three-month journey data from thousands of mobile phone users in the Paris region, France. For the pattern detection, the non-negative matrix factorization (NMF) method is suggested. According to the decompositions of our observation matrix, five types of temporal mobility patterns are obtained. The users’ mobility profiles represented by the impact ratios of these patterns are then interpreted. Beyond these, individual mobility variability is measured correspondingly by the day-to-day and week-to-week mobility divergences. Based on our findings, the impacts of the Covid-19 pandemic and the importance of time-to-time mobility variability are discussed with special concerns on mobility modeling.