Exploiting Unlabeled Data to Enhance Ensemble Diversity