Re-sampling of multi-class imbalanced data using belief function theory and ensemble learning

Imbalanced classification refers to problems in which there are significantly more instances available for some classes than others. Such scenarios require special attention because traditional classifiers tend be biased towards the majority class has a large number of examples. Different strategies, such as re-sampling, have been suggested improve imbalanced learning. Ensemble methods also proven yield promising results presence class-imbalance. However, most them only deal with binary datasets. In this paper, we propose re-sampling approach based on belief function theory and ensemble learning dealing imbalance multi-class setting. This technique assigns soft evidential labels each instance. modeling provides information about object's region, improves selection objects both undersampling oversampling. Our firstly selects ambiguous undersampling, then oversamples minority through generation synthetic examples borderline regions better borders. Finally, induced results, proposed is incorporated into an classifier-independent fusion-based ensemble. The comparative study against well-known reveals that our method efficient according G-Mean F1-score measures, independently from chosen classifier.