Classification Trees for Imbalanced and Sparse Data: Surface-to-Volume Regularization