中文摘要：

针对训练集中类标号存在噪声的情况,提高分类模型的稳定性和分类精度是分类建模的目标。文章通过随机化邻域属性约简,生成多个邻域可分子空间,从而形成不同的基分类模型;通过基分类模型的预测结果及一致性原则学习基分类模型权重,降低了噪声对基分类模型权重学习的影响;最后利用模型权重融合基分类模型的分类结果获得测试样本的类别,并通过仿真实验验证该方法的有效性。

英文摘要：

In view of the class label noise in training dataset, the objective of classification modeling is to improve the stability and classification accuracy of classification model. In this paper, a set of neighborhood separable subspaces is generated based on randomized neighborhood attribute reduction, in which a set of base classification models is obtained. The weight of base classification model is studied by the prediction of base classification model and consensus principle, which decreases the impact of noise data on the weight study of base classification model. Finally, the classification result is gotten by combing the classification decision of different base classification models using model weight, and the experimental results show the validity of the method.