应用于心电图分类的KNNSVM分类器研讨.pdfVIP

2006中国控制与决策学术年会论文集 ChineseControlandDecision 401 of2006 Confernce Proceedings 应用于心电图分类的KNN—SVM分类器研究 谢秋玲 (上海工程技术大学基础学院，上海200336) 攮薹：SVM利用幸近边界的少教向It构造一个最优分类赶平面．在训蝽分类暴时。Fm的着眼点在于两类的麦 鼻部分，那些混杂在另一共中的点往往无助于提高分类嚣的性能，反面套大大增加训蛛器的负担，同时它们逆可生造 成过学习，使迁也生力藏弱．为此，提出了一种改进的舟宪毒算法——KNN—sVM：它先对讽筹袅进行侈剪，根据每十 样奉与其曩近部的K十类标曲异同决定其取音}姑片再甩SVM谢练得封舟夷超平面．实验表明，KNN-SVM相}匕 SVM以及KNN在分类正确率、分类速度以及适用的样丰规模上都表现出了一定的优越性． 关键词，模置识嬲；支将向量机IK近邻舟亮法；棱函薮IVC肇 Classifier toECGData ResearchonKNN-SVM Applying Classification XIE Qiu—ling of China． ofFundamental and 200336t (Department Studies，ShanghaiUniversityEngineeringScience．Shanghai E—mail{qiuling—xie@hotmail．corn) ’ inanother AbstractlSVMfocusthe flearthe in time，andthose intermixed on samples boundarytraining samples of no theclassifiers instead i11creetheburden ctB器are tO performancettheymay usuallygoodimprove greatly and the orderto andtheirexistenceleadto decrease improve computation may over-|itting generalizationability．1n first the OFdeletest the an SVMis set，reserves generalization presented：KNN-SVM．Itprunestraining improved