利用微簇动态欠采样的不平衡数据集成分类算法OA

Ensemble undersampling serves as an effective approach to address class imbalance.However,the perfor-mance of some methods is often compromised due to the loss of key information from the majority class or the disruption of its internal distribution structure.To fully retain the distribution structure of the majority class and enhance the classifi-cation accuracy of the minority class,a Boosting-based ensemble algorithm for imbalanced data is proposed,which dynami-cally samples from majority class micro-clusters.The algorithm constructs majority class micro-clusters using natural nearest neighbors,adaptively determines the sampling size for each micro-cluster based on the sample distribution,and then selects majority class samples according to sampling weights.These selected samples are combined with minority class samples to train an initial base classifier.During subsequent Boosting iterations,the algorithm updates the micro-cluster sampling configuration using classification results from the preceding round.This process reselects majority samples to train new base classifiers,which are finally aggregated into a strong classifier through weighted integration.Comparative experiments conducted on 22 datasets against four classical undersampling methods(Random Undersampling,Cluster Centroids,etc.)and eight mainstream ensemble undersampling methods(Self-paced Ensemble,CusBoost,Equalization Ensemble,etc.)demonstrate that the proposed method achieves superior performance in terms of F1-score,G-mean,and AUC.