Machine learning for credit scoring: Improving logistic regression with non-linear decision-tree effects