An online spatio-temporal model for inference and predictions of taxi demand
Hong Yan, Zhongqiang Zhang, Jian Zou · 2017
Rapid urbanization process has worsened the urban transportation problems such as severe traffic congestion and accidents, and caused significant public safety concerns. The analysis of massive traffic trajectory data is imperative in public transportation surveillance and intelligent transportation. In this work, we analyze taxi calls using the New York City Yellow Cab taxi data in 2015 and propose a Bayesian hierarchical semiparametric model to predict future demands based on the current data. In our hierarchical model, we combine the Dirichlet process and particle filters for the spatio-temporal data analysis. We first partition the region and then employ the space-time model using a stick-breaking construction of the Dirichlet process. Prediction of future demands is carried out using both linear and nonlinear filters. We utilize the cloud computing environment and implement our statistical analysis on a c3.8xlarge Ubuntu Amazon EC2 instance. Finally, we compare the prediction results with those from a Dirichlet process mixture model for the New York City taxi dataset. Our models show advantages in prediction accuracy and computational performance.