基于谱聚类欠取样的不均衡数据SVM分类算法-控制与决策.PDF

第27 卷 第12 期 控 制 与 决 策 2012 年 12 月 Vol. 27 No. 12 Control and Decision Dec. 2012 文章编号: 1001-0920 (2012) 12-1761-08 基于谱聚类欠取样的不均衡数据SVM 分类算法 陶新民, 张冬雪, 郝思媛, 付丹丹 (哈尔滨工程大学信息与通信工程学院，哈尔滨150001) 摘 要: 提出一种基于谱聚类欠取样的不均衡数据支持向量机(SVM) 分类算法. 该算法首先在核空间中对多数类 样本进行谱聚类; 然后在每个聚类中根据聚类大小和该聚类与少数类样本间的距离, 选择具有代表意义的信息点; 最 终实现训练样本间的数目均衡. 实验中将该算法同其他不均衡数据预处理方法相比较, 结果表明该算法不仅能有效 提高SVM 算法对少数类的分类性能, 而且总体分类性能及运行效率都有明显提高. 关键词: 不均衡数据；SVM 算法；谱聚类；欠取样 中图分类号: TP391 文献标志码: A SVM classifier for unbalanced data based on spectrum cluster-based under-sampling approaches TAO Xin-min, ZHANG Dong-xue, HAO Si-yuan, FU Dan-dan (College of Information and Communication Engineering，Harbin Engineering University ，Harbin 150001 ，China. Correspondent ：TAO Xin-min ，E-mail ：taoxinmin@) Abstract: An under-sampling unbalanced dataset support vector machine(SVM) algorithm based on spectrum cluster is presented. Majority instances are clustered by using spectrum cluster in kernel space for resampling reprentative samples with cluster information. The number of selected samples in each cluster is dependent on the size of each cluster and the distance of the cluster to the all minority instances, which can not only reduce the number of majority instances, but also the SVM classification performance under unbalanced dataset is improved by using the proposed method. In the experiments, the proposed approach is compared with other data-preprocess methods for unbalanced dataset classification. The experimental results show that the proposed method can not only improve classification performance of SVM algorithm in the minority class data, but also increase the overall classification performance and effectivity.