基于模糊支持向量机的非平衡数据分类  

Imbalanced data classification based on FSVM

作　　者：陈辉辉[1] 白治江[1] 

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《微型机与应用》2017年第16期56-59,共4页Microcomputer & Its Applications

摘　　要：支持向量机(SVM)作为一种有效的机器学习技术可以很好地处理平衡数据集,然而除了对噪声点和野点敏感以外,SVM在非平衡数据分类时会偏向多数类(负类)样本,从而导致少数类(正类)的分类精度变差。为了克服以上问题,提出了一种改进的模糊支持向量机(FSVM)算法。新算法在设计模糊隶属度时,不仅考虑样本到其所在类中心的距离,还考虑了样本的紧密度特征。实验结果表明,相对于标准SVM及已有的FSVM模型,新方法对于非平衡且含有噪声的数据集有更好的分类效果。As an effective machine learning technology, s u p p o r t ve c to r ma chin e （ SVM） can e f fe c t iv e ly h an d le th e b a la n c c d datasets. H ow e v -er, aside from being sensitive to the noise points and outliers, S VM tends to bias towards the ma jo r i tyl n e g a t iv e ） class in an imb a la n c e d dataset and this leads to a poor classification accuracy of minority! positive） class. In this paper, an im p ro ve d fuzzy s u p p o r t ve c to r m a chin e （ FS-VM ）algorithm is proposed to deal with these problems. When designing the fuzzy membership in the new algoritlim,we take into consideration not only the distance from the sample to the center of its class but also the tightness of the samples. The experimental results show that com-pared to the standard SVM algorithm and the other FSVM models , the new methiod has beter performance in the imbala n ce d and n o is e -c o n ta i-ning datasets.

关 键 词：非平衡数据集 模糊支持向量机 模糊隶属度 样本紧密度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]