When Taxi Meets Bus: Night Bus Stop Planning Over Large-Scale Traffic Data

With more and more citizens traveling for life or work at night, there is a big gap between the demands and supplies for public transportation service in China. In this paper, we address the problem of night-bus stop planning by investigating the characteristics of taxi GPS trajectories and transactions, rather than leveraging subjective and costly surveys about citizen mobility patterns. There are two stages in our method. In the first stage, we extract the Pick-up and Drop-off Records (PDRs) from the taxi GPS trajectories and transactions for capturing citizens travel patterns at night. In the second stage, we propose DC-DBSCAN, an improved DBSCAN clustering algorithm by considering the Distance Constraint, to detect hot locations as candidate night-bus stops from the PDRs dataset. We take the service range of a bus stop into consideration, and optimize the candidates by considering the cost and convenience factors. Finally, our experiments demonstrate that our method is valid and with better performance than that of K-means.