摘要
本文提出了一种新的基于SVM多类问题的策略Half-Against-Half,用该方法训练的基本思想是从多个类别中选择相近或相似的类别,相近的类别放在一个子集里,把多个类别分成两个子集,一直递归地使用这种思想,用类似决策树的思想构造,直到通过多个二分SVM分类器能把每个类别分开。从理论上看,该方法在训练时间、速度、训练集大小等方面比传统的方法OVA、OVO、DAG有一定的优势,并在实践方面得到了实验数据的支持。
A Half-Against-Half (HAH) multi-class SVM is proposed in this paper. HAH is built via recursively dividing the training dataset of K classes into two subsets of classes. The structure of HAH is the same as a decision tree with each node as a binary SVM classifier that tells a testing sample to belong to a certain group of classes. Unlike the commonly used One-Agalnst-All (OVA) and One-Against-One (OVO) implementation methods, both the theoretical estimation and experimental results show that HAH is superior to OVA,OVO and DAG, and is supported by the experimental data in practice, especially in the training time, speed, and training set sizes.
出处
《计算机工程与科学》
CSCD
2007年第4期108-110,共3页
Computer Engineering & Science
基金
湖北省科技攻关计划资助项目(2003BDSP004)