Bike usage forecasting for optimal rebalancing operations in bike-sharing systems

This article presents the first step of a project focusing on enhancing the management of bike-sharing systems. The objective of the project is to optimize the daily rebalancing operations that need to be performed by operators of bike-sharing systems using machine-learning algorithms and constraint programming. This study presents an evaluation of machine learning algorithms developed for forecasting the availability of bikes on three Swiss bike-sharing networks. The results demonstrate the superiority of the Multi-Layer Perceptron algorithm for forecasting available bikes at station-level for different prediction horizons and its applicability for real-time prediction generation.