Abstract: It is difficult for support vector machine to deal with uncertain information because SVM is not only sensitive to noises and outliers but also the inconsistence between conditional features and decision labels is not taken into account. In order to overcome the problem, two types of membership are introduced into standard support vector machine, one type of membership is computed by the distance between the training samples and their center as fuzzy membership, the other type of membership is computed by the distance between the training samples and the nearest training sample with different class label as rough membership. At last several comparative experiments are made to show the performance and the validity of the proposed approach.

Key words: support vector machine, uncertain problem, fuzzy theory, rough set

摘要： 针对支持向量机方法处理不确定信息系统时存在的两个问题：一方面支持向量机训练对噪声样本敏感，另一方面支持向量机训练未考虑信息系统的不一致，利用模糊理论与粗糙集方法分别计算得到两种隶属度：模糊隶属度与粗糙隶属度，并将两种隶属度引入到标准支持向量机中得到一个新的支持向量机模型——双隶属度模糊粗糙支持向量机（DM-FRSVM）。分析该模型对于不确定问题的解决思路并进行对比研究，实验结果表明，在对于含有不确定信息的样本集进行分类时，DM-FRSVM表现出更好的推广性能。

关键词: 支持向量机, 不确定问题, 模糊理论, 粗糙集