Rule induction for global explanation of trained models