Cluster based oversampling for imbalanced learning

Oversampling is a widespread remedy used when there is data imbalance in classification problems. Some oversampling techniques amount to generating new cases in the minority class which are similar to the observed ones. ROSE (Random OverSampling Examples) is an algorithm for generating new data, both in minority and majority classes, by using ideas from kernel density estimation and bootstrap resampling. In this paper, we show that a new strategy which couples density-based clustering methods with ROSE can improve the performance of supervised classification methods with data imbalance. Evidence from some simulation experiments shows that the new procedure is promising and solves some issues related to the use of ROSE.