Minority identification for imbalanced dataset

Minority identification is an important issue in network security and financial applications. This paper considers the direct maximum reachability distance of an object and the indirect minimum reachability distance of an object for measuring the degree of an object being minority. The data classification is performed by an optimized combination model. We empirically evaluate the proposed approach using a number of UCI data sets, and experiment results demonstrate that our method outperforms the existing methods in terms of the comparisons of ROC curves.