Classifying biomedical text abstracts for binary and multi-lass support vector machine using balanced and unbalanced data