Analysis and prediction of bikesharing traffic flow – Citi Bike, New York

Bikesharing systems have witnessed unprecedented growth and significant scholarly attention in recent years. Technological advancement, environmental awareness, and demand for socially equitable transport modes were the major contributors to this development. However, with the ongoing expansion of these systems, companies are faced with the constant need to rebalance them in order to meet the growing demand. Operating companies are continuously searching for more effective and efficient tools for bikesharing traffic flow prediction. This research explores four different techniques for the traffic flow prediction of bikesharing traffic systems including three machine learning algorithms and a statistical time series model. The techniques were evaluated based on prediction accuracy and the best performing algorithm was identified and proposed. In addition, the study analysed the relationship between bike sharing utilisation, weather, and characteristics of bike users, and addressed the neglected aspect of multiple seasonality in time series models. The comparative results confirm that neural networks deliver the best performance. The research evidence suggests that complex seasonalities should be taken into account in traditional time series models.