粗糙one-class支持向量机  

Rough Set-based One-class Support Vector Machine

作　　者：王磊[1,2] 杨一帆[1] 周启海[1] 

机构地区：[1]西南财经大学经济信息工程学院,成都610074 [2]西南财经大学中国支付体系研究中心,成都610074

出　　处：《计算机科学》2009年第9期242-245,共4页Computer Science

基　　金：国家自然科学基金青年项目(60803106);西南财经大学科学研究基金(QN0806)资助

摘　　要：粗糙集理论是处理不确定性和不完备信息的重要方法之一。通过将粗糙集理论引入到one-class支持向量机,提出了一种新颖的粗糙one-class支持向量机。通过定义上近似超平面和下近似超平面,使得训练样本能根据在粗糙间隔中的位置,自适应地对决策超平面产生影响。并且,outlier样本由于距离上近似超平面较近并产生较小的间隔误差,不会导致决策超平面对它们产生明显的过拟合。实验结果表明,粗糙one-class支持向量机的泛化性能优异,识别率和误识率均优于传统的one-class支持向量机。The rough set theory is an important mathematical tool to deal with uncertainty and incompleteness. This paper proposed a novel rough one-class support vector machine by introducing rough margin into one-class support vector machine. With the definitions of upper approximation and lower approximation hyperplanes, the influences of training samples on the decision hyperplane are determined adaptively by their position within the rough margin. Moreover,outlier samples are prone to produce small margin errors since they lie close to the upper approximation hyperplane, so that the overfitting problem of decision hyperplane can be avoided. Experimental results on UCI datasets show the superior generalization performance of rough one-class support vector machine.

关 键 词：粗糙集 ONE-CLASS 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]