基于线性规划的多类支持向量机算法  

A Multi-class Support Vector Machine Algorithm Based on Linear Programming

作　　者：孙德山[1] 吴今培[2] 

机构地区：[1]辽宁师范大学数学学院,大连116029 [2]五邑大学智能技术与系统研究所,江门529020

出　　处：《计算机科学》2005年第10期160-163,共4页Computer Science

基　　金：国家自然科学基金(70471074);广东省自然科学基金(032353)

摘　　要：多类支持向量机一般采用多个两类分类支持向量机来求解,这就需要解多个二次规划问题,从而导致算法的计算复杂性很高。根据一类分类思想,提出一种基于线性规划的多类分类算法及其分解形式,所给算法通过引入核函数能够独立地对每一类样本形成一个紧致的优化区域,从而达到分类的目的。对人工三螺旋线数据和几组实际数据库的识别实验表明,所结算法在保持良好的分类精度前提下,能有效地降低程序的运行时间。The multi-class support vector machine is commonly solved by decomposition to several binary support vector machines, which can bring complicated computation due to solving many quadratic programming problems. A multiclass classification algorithm and its decomposition form based on linear programming are proposed according to oneclass classification idea in this paper, which can form a compact boundary about every single class sample by using kernel function and accordingly obtain the aim of classification. Simulations are conducted on artificial three spiral data and several real databases, which show that the proposed method can reduce the running time of program and guarantee good classification precisior.

关 键 词：线性规划 多类分类 一类分类 核函数 支持向量机算法 线性规划 分类思想 计算复杂性 规划问题 分解形式 分类算法 分类精度 运行时间 

分 类 号：TP301.5[自动化与计算机技术—计算机系统结构] O6-04[自动化与计算机技术—计算机科学与技术]