不同误分代价的支持向量机分类器  

Support Vector Machine with Different Penalty Coefficients

作　　者：业宁[1] 苗哲[1] 刘晓峰[1] 张训华[1] 

机构地区：[1]南京林业大学信息科学技术学院,江苏南京210037

出　　处：《江南大学学报（自然科学版）》2007年第6期658-661,共4页Joural of Jiangnan University (Natural Science Edition)　

基　　金：国家自然科学基金项目(30671639);江苏省自然科学基金项目(BK2005134)

摘　　要：提出一种区分不同类别误分代价的支持向量机模型,建立了拉格朗日方程,利用KKT条件推导出其对偶形式.该模型能够让用户标记不同类别样本的重要性,并根据重要性程度设定不同的误分代价,使重要类别的样本尽量减少被误分的可能性,以满足特定任务的需求.实验结果表明,该模型可以很好地完成特定任务的分类.A support vector machine model that distinguishes different samples category＇s penalty coefficients is proposed. The Lagrange equation is built and its dual form of modal is deduced by KKT condition. This model could let users mark the importance of different samples, and give relevant penalty coefficients according to different importance. This situation can decrease the possibility of being misclassified in important samples. The result of experimentation shows that this model can classify customizing task well.

关 键 词：支持向量机 误分代价 惩罚系数 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]