A Review on Unbalanced Data Classification

Abstract

Classi cation is a supervised machine learning technique to categorize data into a prede ned and distinct number of classes. Again, in the real world, most of these data set are unbalanced. If one of its classes contains signi cantly fewer samples than other classes, this class is called minority class and this data-set is called the unbalanced data-set. The imbalanced property of the data set highly in uenced the performance of traditional classi cation techniques, and classi ers become biased toward the majority class. For the classi cation of an unbalanced data-set, dif- ferent machine-learning techniques are presented by various researchers. In this paper, an attempt is made to summarize popular ML classi ca- tion techniques to handle an unbalanced data set. This paper classi es the existing techniques into three groups: data level approach, algorithm level approach, and classi er's ensemble. This paper also discusses the brief technical details, advantages and disadvantages of these methods. Finally, some of the popular unbalanced data sets available on the UCI repository are also summarized.