Physics-driven machine learning for the prediction of coronal mass ejections' travel times