Exploring App-Based Taxi Movement Patterns from Large-Scale Geolocation Data

This study is designed to leverage ubiquitous mobile computing techniques on exploring app-based taxi movement patterns in large cities. To study patterns at different scales, we comprehensively explore both occupied and unoccupied vehicle movement characteristics through not only individual trips but also their aggregations. Moran's I and its variations are applied to explore spatial autocorrelations among different rides. PageRank centrality is applied for a functional network representing traffic flows to discover places of interest. Gyration radius measures the scope of passenger mobility and driver passenger searching. Moreover, cumulative distribution and data visualization techniques are adopted for trip level characteristics and features analysis. The results indicate that the app-based taxi services are serving more neighborhoods other than downtown areas by taking large proportion of relatively shorter trips and contributing to net increase in total taxi ridership although net decrease in downtown areas. The spatial autocorrelations are significant not only within each service but also among services. With the smartphone-based applications, app-based taxi services are able to search passengers in a larger area and move more efficiently during both occupied and unoccupied periods. Mining from huge empty trip trajectory by app-based taxis, we also identify the existence of stationary/stops state and circulations.