A Cost-Sensitive Ensemble Method for Class-Imbalanced Datasets