Exploratory review on class imbalance problem: An overview

Nothing is as important as human lives. There are so many cases where we need to predict the things, the causes that lead to the destruction of mankind. Such things occur occasionally but can be destructive. Therefore their prediction is very important so as to solve them at the early and safe stage. Such data is said to be imbalanced where negative cases outnumber the positive cases by huge proportions and the prediction of these rare occurring positive cases is very important. So far all the machine learners are biased towards the majority class. In this overview we are exploring all the techniques that have been used to mine the imbalanced data sets. Techniques at pre-processing level, algorithmic level are being discussed in this review. Also ensemble and hybrid techniques are being reviewed. In this paper, techniques of two types of imbalanced data sets are being reviewed viz binary class imbalanced data and multi class imbalanced data.