树形层次结构的非平衡SVM分类方法

Imbalanced SVM classification algorithm based on tree hierarchical structure

针对支持向量机(support vector machine,SVM)无法对非平衡数据有效分类的问题,提出树形层次结构的非平衡SVM(imbalanced SVM method based on tree hierarchical structure,ISVM_TH)分类方法。通过衡量多数类样本与超平面之间的关系,有效区分不同类的重要性,提取关键簇,通过对关键簇进行逐层划分,构建更为合理的多数类样本树形层次结构,提取候选支持向量(candidate support vector,CSV)信息,参与SVM的训练过程,提高SVM对于非平衡数据的分类能力。实验结果表明,该方法能够有效改善SVM对于非平衡数据的分类性能,获得令人满意的泛化能力。

To solve the problem that the traditional SVM cannot solve the imbalanced datasets classification problems efficiently, an imbalanced support vector machine （SVM） based on tree hierarchical structure （ISVM_ TH） classification algorithm was proposed. The key clusters were extracted and the importance of various clusters was distinguished by measuring the relationship between the majority class samples and the hyperplane. And the more reasonable tree hierarchical structure for majority class samples was constructed through the layer by layer division for the importance clusters. The candidate support vectors （CSV） were extracted and trained using SVM to improve the classification ability of SVM for imbalanced datasets. Experimental results demonstrate that the proposed method can improve the classification results for imbalanced datasets classification problems and obtain good gene-ralization performance.