Handling imbalanced data with decision trees