可补偿类别差异的加权支持向量机算法

支持向量机(SVM)算法在各类别样本数多少不同时,样本数量多的类别,其分类误差小,而样本数量少的类别,其分类误差大.针对这种倾向性问题,在分析其产生原因的基础上,提出了加权SVM算法,从而克服了常规SVM算法不能灵活处理每一个样本的缺陷,同时补偿了这种倾向性造成的不利影响.这种以牺牲大类别精度来提高小类别精度的加权支持向量机方法,可应用于诸如故障诊断等关注小类别分类精度的场合.户外图象识别的实验结果证明,该算法是有效的.

Weighted Support Vector Machine Based Classification Algorithm for Uneven Class Size Problems

When training sets with uneven class sizes are used, the classification result based on support vector machine (SVM) is undesirably biased towards the class with more samples in the training set. That is to say, the larger the sample size, the smaller the classification error, whereas the smaller the sample size, the larger the classification error. This paper proposes weighted support vector machine algorithms based on the analysis of the cause of such problem, and this algorithm overcomes the drawback which standard support vector machine algorithm can't deal with each sample flexibly and compensates for the unfavorable impact caused by this bias. Such weighted support vector machines improve classification accuracy for class with small size at the cost of accuracy reduction for large size class, and can be applied to the case of regarding small sort of classification accuracy, such as fault diagnosis. The result of outdoor image recognition shows the effectiveness of this algorithm.