Prediction of Bus Arrival Time Using Real-Time on-Line Bus Locations

Abstract

Reliable and accurate prediction of bus arrival time is considered as one of the important services to attract people’s choice of bus ridership. In this paper, we develop a simple yet accurate real-time bus arrival prediction system for a crowded small tourist city, like Macao, using accurate on-line bus locations provided by the government website. These accurate bus locations are freely available on-line, which are generated by dedicated sensors installed in bus stops and buses. We first proposed a link time model for storing all of the link times between adjacent bus stops on different bus routes in the entire network so that the trip time between any two of the bus stops in the network can be predicted in real-time. Three different simple models based on historical and real-time on-line bus locations, namely Simple Moving Average (SMA), Artificial Neural Network (ANN) and Hybrid Model, are proposed for the bus arrival prediction system, taking into account the real-time weather conditions. It was found that the Hybrid model perform the best among the three models. The average mean absolute percentage error (MAPE) for the Hybrid model is 17% and the average mean absolute error (MAE) and root mean square error (RMSE) is less than 1 minute. For future works, more advanced deep learning models with Kalman filtering can be evaluated, using on-line bus locations from more bus routes.