Bus Arrival Time Estimation for Public Transportation System Using LSTM

Abstract

This research aims to create an arrival time estimation model for the electric bus service of Chiang Mai University. To achieve that, we employed the Long Short-Term Memory (LSTM) model to capture the regularities in the data. The model was trained using online learning approach well suited for learning from real-time data coming directly from the bus's onboard GPS unit. Despite the size and the complexity of the data, LSTM demonstrated the capability to learn from the data. Experimental results based on real data records of four representative bus lines, spanning over 3 months demonstrated the superiority of the proposed LSTM compared to the Support Vector Regression (SVR) as measured by the Mean Absolute Error (MAE). This suggested that the estimation method proposed in this study is feasible and might be applicable to other bus service networks.