摘要
层次支持向量机(SVM)是多类分类方法应用中的研究热点。针对SVM的分类面仅由支持向量决定的理论,提出一种基于无监督聚类方法来预抽取支持向量,训练向量机;并分析现有多类分类方法所存在的弊端,基于综合考虑节点的类集合可分性,设计一种基于树分类器整体性能最优的SVM二叉树层次分类方法。实验表明,该方法对比传统一类对余类法和成对分类法在整体分类精度和训练时间上都有明显提高。
Hierachical support vector machine （SVM） is a study hotpoint in application of the multi-classification method. The main purpose of this paper is to propose a unsupervised clustering-based method to pre-extract the support vector for training the SVM in light of the theory that the classification facet of SVM is determined by support vector only. By analyzing the limitation existed in current multi-class classification methods, and taking into account comprehensively the separability of the class set of nodes, a hierarchical classification method of SVM binary tree based on tree classifier with optimized overall performance is designed. Experimental results indicate that comparing with traditional method of one versus rest（OVR） and the method of one versus one （OVO） , the new method has noticeable enhancement in overall classification precision and training time.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第12期226-228,共3页
Computer Applications and Software
