一种基于属性加权的代价敏感支持向量机算法

针对实际中存在的各类别样本错分造成不同危害程度的分类问题,提出了一种基于属性加权的代价敏感支持向量机分类算法,即在计算各个样本特征属性对分类的重要度之后,对相应的属性进行重要度加权,所得的数据用于训练和测试代价敏感支持向量机。数值实验的结果表明,该方法提高了误分代价高的类别的分类精度,同时属性重要度的引入提高了分类器的整体分类性能。该方法对错分代价不对称的数据分类问题具有重要的现实意义。

Cost-Sellsitive support vector machine based on weighted attribute

In practice it was existed different wrong-classified-cost matter in classified problem. This paper proposed a cost-sensitive SVM approach based on weighted attribute, which firstly calculated the weightiness of feature attributes corresponded to the classification attribute, then calculated the corresponding weightiness of attribute for all samples, finally the samples were used for cost-sensitive SVM training and testing. The experimental results showed that the approach can improve the classification precision of the costsensitive samples, and the use of feature attribute increased the integer classified capability of the classifier. The approach has important realistic significance of unbalanced wrong-classification cost in classified problems.