This paper addresses the class imbalance problem in machine learning, which is common in many domains such as text classification, target detection, and fraud detection. The authors explore the effectiveness of different resampling methods, including oversampling and undersampling, and their combinations. They conduct an experimental study on various datasets to determine the optimal resampling strategies and rates. The results show that combining different resampling approaches can improve classification accuracy compared to using a single method. The proposed combination scheme is evaluated on imbalanced subsets of the Reuters-21578 text collection and is found to be effective. The paper also discusses the architecture and implementation details of the combination scheme, which biases classifiers towards the underrepresented class and uses an elimination process to prevent unreliable classifiers from participating in decision-making. The proposed method outperforms both C4.5 and Adaboost in terms of classification accuracy on various datasets.This paper addresses the class imbalance problem in machine learning, which is common in many domains such as text classification, target detection, and fraud detection. The authors explore the effectiveness of different resampling methods, including oversampling and undersampling, and their combinations. They conduct an experimental study on various datasets to determine the optimal resampling strategies and rates. The results show that combining different resampling approaches can improve classification accuracy compared to using a single method. The proposed combination scheme is evaluated on imbalanced subsets of the Reuters-21578 text collection and is found to be effective. The paper also discusses the architecture and implementation details of the combination scheme, which biases classifiers towards the underrepresented class and uses an elimination process to prevent unreliable classifiers from participating in decision-making. The proposed method outperforms both C4.5 and Adaboost in terms of classification accuracy on various datasets.