Public transport, especially the bus transport, can reduce the private car usage and fuel consumption, and alleviate traffic congestion. However, when traveling with buses, the travelers not only care about the waiting time, but also care about the crowdedness in the bus itself. Excessively overcrowded bus may drive away the anxious travelers and make them reluctant to take buses. So accurate, real-time and reliable passenger demand prediction becomes necessary, which can help determine the bus headway and reduce the waiting time of passengers. However, there are three major challenges for predicting the passenger demand on bus services: inhomogeneous, seasonal bursty periods and periodicities. To overcome the challenges, we propose three predictive models and further take a data stream ensemble framework to predict the number of passengers. We develop an experiment over a 22-week period. The evaluation results suggest that the proposed method achieves outstanding prediction accuracy among 86,411 passenger demands on bus services, more than 78% of them are accurately forecasted.
