基于鲁棒近邻粗糙逼近的属性约简算法  

Attribute Reduction Algorithm Based on Robust Neighborhood Rough Approximation

作　　者：郑文彬 胡敏杰 何秋红 ZHENG Wenbin;HU Minjie;HE Qiuhong(School of Computer Science,Minnan Normal University,Zhangzhou,Fujian 363000,China;Lab of Granular Computing,Zhangzhou,Fujian 363000,China)

机构地区：[1]闽南师范大学计算机学院,福建漳州363000 [2]福建省粒计算及其应用重点实验室,福建漳州363000

出　　处：《闽南师范大学学报（自然科学版）》2018年第4期22-31,共10页Journal of Minnan Normal University：Natural Science

基　　金：福建省教育厅科技项目(JAT170347;JAT170350)

摘　　要：基于邻域粗糙集的属性约简模型既受邻域半径参数值的影响,又不能评估属性与样本对象之间的内在关系.为此,本文先提出鲁棒近邻来确认对象的邻域,计算出若干个与样本对象最近同类与最近异类对象距离的平均值,然后依据分类区分函数的定义来确定近邻类的邻域半径大小,构造了鲁棒近邻粗糙集模型.最后按照其模型,基于样本对象对属性的评价准则提出了鲁棒近邻的属性约简算法.该算法模型分别在CART,KNN和LSVM三个分类器和10个样本数据集中测试运行,实验效果表明该模型不但可以筛选得到较少的属性集,而且还可以有效提高分类精度.Attribute reduction based on neighborhood rough sets is restricted by the neighborhood radius value,which can also not evaluate relationship between attributes and sample objects.In this paper,a robust neighborhood is presented to estimate sample's neighborhood.The average of the distances between the multiple nearest missing,the multiple nearest hit of a given sample is calculated,then the neighborhood radius of the nearest neighbor according to discernibility function is determined to construct robust neighborhood rough sets model.Finally,according to its model,a robust nearest neighbor attribute reduction algorithm based on the evaluation criterion of sample pair attributes is proposed.The experiment is conducted on CART,KNN and LSVM three different classifiers and ten different datasets.The experimental results show that the model can not only filter fewer attribute set,but also improve the classification accuracy effectively.

关 键 词：属性约简 鲁棒近邻 区分函数 邻域粗糙集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]