Safe level OUPS for improving target concept learning in imbalanced data sets

Binary or two group classification is made difficult when the groups are skewed or imbalanced. This class imbalance will induce bias into the classifier particularly when the imbalance between both groups is high. Binary class imbalance usually suffers from data intrinsic properties beyond that of class imbalance alone. In this paper we discuss these data intrinsic properties that contribute to degradation of classifier performance in class imbalance data sets and introduce a state of the art pre-processing technique that improves concept learning within class imbalanced data. We perform simulation experiments and compare our technique against other popular techniques as well as combinations that have been used to improve classifier performance for imbalanced data sets. The results of the experiments show that the Safe Level OUPS approach outperforms other techniques in regards to sensitivity measures. We discuss and analyze competing techniques and highlight the pros and cons of using these techniques.