Classification of minority class in imbalanced data sets

Abstract

This paper focuses on imbalanced data sets and uses different methodologies for the part of classification process related to building train and test subsets. Popular classification methods are then applied and evaluated based on the recall result for the minority class. On the basis of the results from our experiments we suggest that for imbalanced data sets when the minority class presents noticeably higher interest, we should use alternative methodology for building the train subset and not the standard random allocation of observations, in order to improve the predictive power of the classifier for the minority class.