基于三分类支持向量机的多分类算法的研究

Study on Classification Algorithm Based on Three Classification Support Vector Machine

用支持向量机解决多分类问题是目前众多学者研究的热点话题.将已有的最小二乘支持向量分类-回归机算法推广到M空间进行了理论分析,在基于支持向量机的三分类算法基础上,提出了两个新的K(K>3)类多分类算法:一对一对多与一对一对一算法.对所有数据集进行分类时,在已有的多分类算法的基础上采用加校正的技巧:忽略准确率低的子分类器.数值实验证明了该技巧的有效性,并且校正后的准确率比校正前平均提高了4.61%.

Solving multi-class classification problems by means of support vector machine is a hot topic researched by many scholars.Firstly,Least Square Support Vector Classification-Regression algorithm was analyzed in the M space.Then two new multi-classification algorithms for Kclasses：1-v-1-v-r（one versus one versus rest）and 1-v-1-v-1（one versus one versus one）were proposed based on tri-classification algorithm.When all of the data set were classified,adding correction and ignoring the low accuracy of the sub classifiers were used,which based on the existing multi-classification algorithm.The numerical experiments show that the algorithm is effective and the accuracy is improved by 4.61% after the correction.