Discovering outlier filtering rules from unlabeled data