Decision Boundary Computation-based Over-sampling for Imbalance Learning

Over-sampling is a very effective method to solve
the imbalanced problem by generating new synthetic samples for the minority class. But rare over-sampling methods focus on the borderline between classes and only use the linearinterpolation between boundary samples to fill the decision boundary, so not take full use of information in the decision boundary at all. To fill this gap, one novel method named Decision Boundary Computation-based Oversampling is proposed. Firsts, the novel method treats surrounding areas of both boundary majority and minority samples as the decision boundary. Then, compute it’s area belonging to majority class and treat the remained one as the area belonging to the minority class. Thus, the novel method greatly enhances the full use of boundary information and implicitly complements the nature insufficiency of information of minority class at the same time. Finally, new synthetic samples are generated in the partition of decision boundary of minority class. Extensive experiments indicate the good performance of proposed method when compared with other state-of-art methods.