电子学报 ›› 2002, Vol. 30 ›› Issue (5): 745-748.

• 论文 • 上一篇    下一篇

SVM-KNN分类器——一种提高SVM分类精度的新方法

李 蓉, 叶世伟, 史忠植   

  1. 1.中国科技大学研究生院(北京)计算机教学部,北京 100039;2.中国科学院计算技术研究所智能信息处理实验室,北京 100080
  • 收稿日期:2001-06-15 修回日期:2001-10-06 出版日期:2002-05-25

SVM-KNN Classifier——A New Method of Improving the Accuracy of SVM Classifier

LI Rong, YE Shi-wei, SHI Zhong-zhi   

  1. 1.Dept.of Computing,Graduate School,Science and Technology University of China,Beijing 100039,China;2.National Key Laboratory of Intelligent Information Technology Process,The Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100080,China
  • Received:2001-06-15 Revised:2001-10-06 Online:2002-05-25 Published:2002-05-25

摘要: 本文提出了一种将支持向量机分类和最近邻分类相结合的方法,形成了一种新的分类器.首先对支持向量机进行分析可以看出它作为分类器实际相当于每类只选一个代表点的最近邻分类器,同时在对支持向量机分类时出错样本点的分布进行研究的基础上,在分类阶段计算待识别样本和最优分类超平面的距离,如果距离差大于给定阈值直接应用支持向量机分类,否则代入以每类的所有的支持向量作为代表点的K近邻分类.数值实验证明了使用支持向量机结合最近邻分类的分类器分类比单独使用支持向量机分类具有更高的分类准确率,同时可以较好地解决应用支持向量机分类时核函数参数的选择问题.

关键词: 支持向量机, 最近邻分类, 类代表点, 核函数, 特征空间, VC维

Abstract: A new algorithm that combined Support Vector Machine (SVM) with K Nearest neighbour (KNN) is presented and it comes into being a new classifier.The classifier based on taking SVM as a 1NN classifier in which only one representative point is selected for each class.In the class phase,the algorithm computes the distance from the test sample to the optimal super-plane of SVM in feature space.If the distance is greater than the given threshold,the test sample would be classified on SVM;otherwise,the KNN algorithm will be used.In KNN algorithm,we select every support vector as representative point and compare the distance between the testing sample and every support vector.The testing sample can be classed by finding the k-nearest neighbour of testing sample.The numerical experiments show that the mixed algorithm can not only improve the accuracy compared to sole SVM,but also better solve the problem of selecting the parameter of kernel function for SVM.

Key words: support vector machine, nearst neighbour algorithm, representative point, kernel function, feature space, VC Dimension

中图分类号: