Abstract: Performance of an ensemble classifier depends on its base classifiers and methods of combination，a selective SVM ensemble classifier based on rough set is proposed，which firstly uses attribute reduction based on rough set（RS） to generate a set of reductions and each reduction is used to develop a base classifier.Those base classifiers which satisfy both individual accuracy and diversity are selected and combined by the way of majority voting.The experiment results on UCI database show the performance of this method.

摘要： 集成分类器的性能很大程度决定于各成员分类器的构造和对各成员分类器的组合方法。提出一种基于粗集理论的选择性支持向量机集成算法，该算法首先利用粗集技术产生一个属性约简集合，然后以各约简集为样本属性空间构造各成员分类器，其次通过对各成员分类器精度与差异度的计算，选择既满足个体的精度要求，又满足个体差异性要求的成员分类器进行集成。最后通过对UCI上一组实验数据的测试，证实该方法能够有效提高支持向量机的推广性能。