Bus Arrival Time Prediction for Short-Distance Bus Stops with Real-Time Online Information

It is desirable to provide the accurate bus arrival time for bus riders' convenience and for bus service providers' quality of services. In this paper, we compare and evaluate a realtime bus arrival prediction system for short-distance (< 1km) bus stops using real time and publicly available online information, including bus location, weather, traffic status and passenger flow. We compare two types of prediction methods for predicting a trip time (time between multiple stops). One is to use the sum of link times (time between adjacent stops), while the other is to use the entire trip time to predict the corresponding trip time. We evaluate and compare different prediction models using Support Vector Machine (SVM), Artificial Neural Network (ANN), Long Short- Term Memory (LSTM) and dual-stage attention-based Recurrent Neural Network (DA-RNN) in link time prediction, taking into accounts of the real-time online information. It was found that DA-RNN performs the best among all models considered, with an average link time Mean Absolute Percentage Error (MAPE) of about 18% and trip time MAPE of about 10%. For trip time prediction, the larger distance of a trip (more bus stops) the more accurate the prediction is. DA - RNN model is practical to use for bus arrival time prediction in Macao, a tourist city with millions of tourists and short-distance bus stops.