Predicting station level demand in a bike-sharing system using recurrent neural networks

The modern multi-modal transportation system has revolutionised the landscape of public mobility in cities around the world, with bike-sharing as one of its vital components. One of the critical problems in persuading citizens to commute using the bike-sharing service is the uneven bikes distribution which leads to bike shortage in certain locations, especially during rush hours. This study offers a system, which provides predictions of both rental and return demand in real-time for each bike station by using only one model, which can be used to formulate load balancing strategies between stations. Five different architectures based on recurrent neural network are described and compared with four evaluation metrics: mean absolute percentage error, root mean squared logarithmic error, mean absolute error and root mean squared error. This system has been tested with New York Citi Bike dataset. The evaluation shows the authors' proposed methods demonstrate satisfying results at both the global and station levels.