Data mining with imbalanced class distributions: concepts and methods

Some real world data mining applications present imbalanced or skewed class distributions. In these domains, the underrepresented classes are often the ones we are more interested in. However, most learning algorithms are not able to induce meaningful classifiers in some imbalanced domains. One reason for this poor performance is that learning algorithms tend to focus in abundant classes to maximize classification accuracy. This paper reviews recent work in this subject, focusing in concepts and methods to deal with imbalanced data sets.