基于拆分集成的不均衡数据分类算法

Imbalanced Data Classification Algorithm Based on Split and Classifier Ensemble

为改进SVM对不均衡数据的分类性能,提出一种基于拆分集成的不均衡数据分类算法,该算法对多数类样本依据类别之间的比例通过聚类划分为多个子集,各子集分别与少数类合并成多个训练子集,通过对各训练子集进行学习获得多个分类器,利用WE集成分类器方法对多个分类器进行集成,获得最终分类器,以此改进在不均衡数据下的分类性能.在UCI数据集上的实验结果表明,该算法的有效性,特别是对少数类样本的分类性能.

To improve the performance of Support Vector Machine classifier for imbalanced data, an imbalanced data classification algorithm based on split and classifier ensemble is introduced. The majority class sample is divided into several sub sets by clustering, and each subset is combined with minority class sample to produce a training subset. Then the training subsets are learned and multiple classifiers are obtained. Finally the multiple classifiers are integrated and the ensemble classifier is obtained. Experimental results show the algorithm is effective for imbalanced dataset, especially for the minority class samples.