作　　者： ; ; ;

机构地区： 湘潭大学信息工程学院

出　　处： 《控制与决策》 2010年第3期371-377,共7页

摘　　要： 为了提升集成网络的泛化性能,在Boosting或Bagging算法对样本进行扰动的基础上,通过粗糙集约简实现特征属性选择,将样本扰动和输入属性扰动结合起来,提出了Rough_Boosting和Rough_Bagging算法.该算法通过生成精确度高且差异度大的个体网络,提高了集成的泛化能力.实验结果表明,该算法泛化能力明显优于Boosting和Bagging算法,且生成的个体网络差异度更大,与同类算法相比,具有相近或相当的性能. Rough_Boosting and Rough_Bagging algorithms are proposed to improve the generalization ability of ensemble networks. The training samples are disturbed by using conventional Boosting or Bagging algorithm. Proper attribute reducts are selected by using rough sets theory. Thus,the mechanism of disturbing samples and input attributes are combined. The generalization ability is improved by generating accurate and diverse component networks. Experiment results show that the generalization ability of proposed method is obviously better than that of Boosting and Bagging methods,and individual networks are more diverse. Compared with prevailing similar methods,the method has close or approximate performance.

关 键 词： 粗糙集 约简 神经网络集成 特征选择

领　　域： [自动化与计算机技术] [自动化与计算机技术]