Modeling and Analysis of Daily Driving Patterns of Taxis in Reshuffled Ride-Hailing Service Market

This paper aims to model and analyze the changes in daily driving patterns of taxis in a disrupted market due to the boom in e-hailing services. This is accomplished by mining large-scale trajectory data sets obtained from a major taxi company in Shanghai. The taxi data set includes more than 0.8 billion trajectory points associated with over 12,000 taxis obtained in a period of 10 days (5 continuous weekdays in 2012 and 2016, respectively). The raw data were efficiently processed with the acceleration of high-performance computing. Creatively, the concept of information entropy together with principal component analysis were adopted to spatially delineate the gridded daily taxi driving trajectories. This helps describe the disordered taxi traces in comparable profiles across different spatial zones. Then, distinct patterns were extracted using the k-means clustering method. The proposed analysis pipeline has built a stable way of comparing driving patterns between different time periods after relaxing concerns about potential spreading of demand over time. By comparing statistical features associated with the identified clusters, the changes in daily taxi driving patterns in the context of the wide popularization of e-hailing services were quantitatively unveiled. This will be informative for taxi service providers revamping their business models when facing the opportunities brought by e-hailing apps and competition from other ride-sourcing vehicles in urban areas.