融合密度与邻域覆盖约简的分类方法  

A classification method by integrating density and neighborhood covering reduction

作　　者：张清华[1] 艾志华 张金镇 ZHANG Qinghua;AI Zhihua;ZHANG Jinzhen(Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学计算智能重庆市重点实验室,重庆400065

出　　处：《陕西师范大学学报（自然科学版）》2022年第3期33-42,共10页Journal of Shaanxi Normal University：Natural Science Edition

基　　金：国家重点研发计划(2020YFC2003502);国家自然科学基金(61876201)。

摘　　要：针对现有大部分基于邻域覆盖的分类方法直接根据最近的邻域对样本分类,没有考虑邻域之间的差异性,从而导致分类错误的问题。引入局部密度去刻画邻域之间的差异性,并提出融合密度与邻域覆盖约简的分类方法(D-NCR)。首先,通过邻域覆盖约简剔除冗余的邻域;其次,将邻域中心的局部密度转化为权重,定义了测试样本到邻域中心的加权距离;最后,基于加权距离,对不同的测试样本提出两种分类策略。在UCI数据集上的实验结果表明,所提方法能达到较好的分类效果。For the problem that most existing classification methods based on neighborhood covering directly classify samples according to the nearest neighborhood.The certain classification methods do not consider the differences between each neighborhood,which may lead to classification mistakes.Local density is introduced to characterize the differences between each neighborhood,and a classification method by integrating density and neighborhood covering reduction(D-NCR)is proposed in this paper.First,the redundant neighborhoods are eliminated by the neighborhood covering reduction.Then,the weighted distance from the test sample to the neighborhood center is defined through local density of neighborhood center.Finally,two classification strategies are proposed for different test samples based on the weighted distance.Experimental results on UCI data sets demonstrate the effectiveness of the proposed method.

关 键 词：邻域粗糙集 邻域覆盖约简 局部密度 加权距离 分类策略 

分 类 号：TB18[一般工业技术] O225[理学—运筹学与控制论]