Bus Arrival Time Prediction Using Recurrent Neural Network with LSTM Architecture

Abstract

Arrival time of public vehicles to transport stops is a key point of information systems for passengers. Accurate information on the arrival time is important for travel arrangements since it helps to decrease the wait time at a stop and to choose an optimal alternate route. Recently, such information has been included to mobile navigation applications too. In the present paper, we analyze the abilities of the LSTM neural network to predict the arrival time of public vehicles. This model accounts for heterogeneous information about transport situation that directly or indirectly has an impact on the travel time prediction and includes statistical and real-time data of traffic flow. We examined the model experimentally using traffic data on bus routes in the city of Samara, Russia. The obtained results confirm that the predictions provided by our model are of a high quality and it can be used for real-time arrival time prediction of public transport in the case of a large-scale transportation network.