Machine learning from imbalanced data-sets: An application to the bike-sharing inventory problem