Mining Multiple Class Imbalanced Datasets Using a Specialized Balancing Algorithm and the Adaboost Technique

In this paper, we propose an ensemble classifier extended from a specialized bicriterion balancing algorithm originally proposed by the authors for binary imbalanced classification. The approach uses two specialized criteria for oversampling - classification potential and distance from the borderline between minority and majority instances. For mining multiclass imbalanced datasets the bicriteria oversampling algorithm was adapted to the needs of the multiple class problems using the one-versus-one (OVO) approach and the Adaboost technique. To evaluate the performance of the proposed ensemble classifier we use several state of the art balancing algorithms. The computational experiment shows a very good performance of the proposed approach.