Development of a station-level demand prediction and visualization tool to support bike-sharing systems’ operators

Bike-sharing systems operate in a number of cities around the world, aiming to promote sustainable urban mobility. Successful management of these systems is to a large extent linked to the optimal distribution of bicycles, which implies the accurate prediction of demand for rentals and returns at each station within the day. For this purpose, a tool for predicting bike demand for rentals and returns and visualizing the results has been developed and is presented in the present paper. Different predictive models based on machine learning regression algorithms are trained and evaluated. The tool is tested using data from the bike-sharing system that operates in the city of Thessaloniki, Greece for which the results indicate that the tested system’s utilization is highly correlated to the location and spatial characteristics of a station, as well as the season of the year and time of day. The proposed machine learning algorithms use custom engineered features to learn those correlations and achieve the highest possible performance.