支持向量机多类分类方法

SVM Multi-Class Classification

支持向量机本身是一个两类问题的判别方法,不能直接应用于多类问题。当前针对多类问题的支持向量机分类方法主要有5种:一类对余类法(OVR),一对一法(OVO),二叉树法(BT),纠错输出编码法和有向非循环图法。本文对这些方法进行了简单的介绍,通过对其原理和实现方法的分析,从速度和精度两方面对这些方法的优缺点进行了归纳和总结,给出了比较意见,并通过实验进行了验证,最后提出了一些改进建议。

The support vector machine （SVM） is used for the binary-class classification. It cannot deal with multi-class classification directly. Five methods for multi-class classification are introduced based on widely used SVMs. They are one versus rest（OVR）,one versus one （OVO）, binary tree（BT）,error correcting output codes（ECOC） and directed acyclic graph （DAG）. A comparison result about the classification speed and accuracy is given through theoretic analysis. Experimental results demonstrate the comparison result. In addition, some suggestions for improving these methods are also presented.