Learning on the border: active learning in imbalanced data classification