The problem of classification in imbalanced data sets in knowledge discovery

Abstract

It has been observed that classification in imbalanced data sets have drawn more attention to researchers in knowledge discovery and data mining fields. In such problems, almost all the samples are labeled as one class, while far fewer samples are labeled as the other class, which are usually more important. But traditional classifiers that try to pursue whole accurate performance over a full range of samples are not suitable to deal with classification in imbalanced data sets, since they tend to biases towards majority class while pay less attention to the rare one. In the present work, we perform a review of the most important research lines on this topic and point out several directions for further investigation.