RRBoost: a new ensemble method for classifying imbalanced data

Class imbalance is a common issue in classification tasks, often causing standard classification models to misclassify instances from minority classes. Recent efforts to address this problem frequently involve the combination of sampling techniques with ensemble models. Extending this trend, we propose a new method called R-ROSE Boosting (RRBoost). This method involves the development of a novel synthetic data generation technique, termed radius-random oversampling examples (R-ROSE), and its integration with a boosting-based ensemble method. This approach offers the advantage of enhancing the diversity of synthetic data in the vicinity of hard-to-classify observations, thereby potentially improving classification accuracy. We demonstrate the effectiveness of RRBoost by comparing it with other ensemble models using 24 real imbalanced datasets. As a result, RRBoost proves to be an effective method for addressing imbalanced data, demonstrating superior classification performance compared to other ensemble models.