摘要
针对二叉树多类分类方法存在的问题,提出了一种基于Huffman树的构造过程自下而上生成二叉树结构的方法。为降低二叉树方法"误差累积"的影响,使用模糊支持向量机来训练每个结点的两类分类器。针对设计隶属度函数时仅考虑样本与类别关系,而不考虑样本间关系的问题,提出了根据传统支持向量机构造的超平面做切球来确定样本间关系的方法,有效地区分了有效样本和噪音、孤立点样本。实验结果表明:同其他多类支持向量机方法相比,该方法具有更好的分类性能。
In order to deal with the deficiency of binary tree multiclass classification methods, a method of constructing a binary tree from down to up according to the construction process of Huffman tree is proposed. To reduce the influence of binary tree＇s ＂accumulated errors＂, a fuzzy support vector machine is ased to train a two-class classifier of every binary tree node. In order to deal with the fuzzy membership＇s deficiency of taking the relation between a sample and its cluster center into account, but not those among samples, which is described by the affinity among samples, a method of defining the affinity among samples according to tangent sphere based on traditional support vector machine＇s hyperplane is proposed, and effectively distinguish outliers and noises samples from effective samples. Experimental results show that compared with other multiclass support vector machine methods, the method we proposed has better classification property.
出处
《辽宁工程技术大学学报（自然科学版）》
CAS
北大核心
2008年第1期96-99,共4页
Journal of Liaoning Technical University (Natural Science)
基金
甘肃省科技攻关计划基金资助项目(2GS047-A52-002-03)
