基于监督信息邻域粒距离的属性约简方法  

Attribute reduction method based on supervised information neighborhood granular distance

作　　者：支宇 胡军[1,2] ZHI Yu;HU Jun(School of Computer Science&Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065 [2]重庆邮电大学计算智能重庆市重点实验室,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2023年第5期898-907,共10页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金项目(61876201,61876027);重庆市自然科学基金项目(cstc2019jcyj-cxttX0002,cstc2021ycjh-bgzxm0013);重庆市教委重点合作项目(HZ2021008)。

摘　　要：在邻域粗糙集的研究中,主要通过邻域关系来实现信息的粒化。传统的邻域关系容易引起信息的不一致,导致将确定可分的样本划分到边界域里,使得属性约简的结果不理想。将监督信息(标签信息)引入邻域粒,提出了一种新的属性约简方法。该方法利用样本邻域的监督信息构造出监督信息邻域粒距离,并基于此定义近似正域。将原边界域中确定可分的样本划分到正域中,从而更精准地刻画邻域系统的分类能力。实验结果表明,所提出的方法不仅能够有效删除冗余属性,而且还能提高分类性能,证明了在属性约简中考虑监督信息邻域粒距离的可行性和有效性。In the research of neighborhood rough sets,information granulation is mainly realized through neighborhood relationship.The traditional neighborhood relation is easy to cause the inconsistency of information,leading to the division of definitely separable samples into boundary region,and making the result of attribute reduction unsatisfactory.In this paper,supervised information(label information)is introduced into neighborhood granules,and a new attribute reduction method is proposed.This method uses the supervised information of the sample neighborhood to construct the supervised information neighborhood granular distance,and defines the approximate positive region based on it.In this way,the separable samples in the original boundary region can be divided into the positive region,and the classification ability of neighborhood system can be described more accurately.The experimental results show that the proposed method can not only effectively delete redundant attributes,but also improve the classification performance,proving the feasibility and validity of considering the supervised information neighborhood granular distance in attribute reduction.

关 键 词：粗糙集 邻域关系 边界域 监督信息 属性约简 

分 类 号：TN919[电子电信—通信与信息系统] TP311[电子电信—信息与通信工程]