An approach for classification of highly imbalanced data using weighting and undersampling