三支决策视角下的属性约简加速方法  

An attribute reduction acceleration method based on three-way decisions

作　　者：姜春茂[1] 刘安鹏 JIANG Chun-mao;LIU An-peng(School of Computer Science and Information Engineering,Harbin Normal University,Harbin 150025,China)

机构地区：[1]哈尔滨师范大学计算机科学与信息工程学院,黑龙江哈尔滨150025

出　　处：《计算机工程与科学》2020年第12期2280-2286,共7页Computer Engineering & Science

基　　金：国家自然科学基金(61202458,61403109);黑龙江省自然科学基金(F2017021)。

摘　　要：属性约简是粗糙集领域的核心研究内容,在此基础上发展出了集成属性约简。所谓集成约简是将样本依据决策类别划分为多个决策系统分别计算。集成属性约简虽然能够平衡各个决策类的需求,但增加了约简的时间消耗。为了解决这一问题,提出了一种基于序贯三支决策的属性约简加速方法。具体步骤如下:(1)计算决策系统中的属性重要度;(2)将属性重要度的结果进行三分,重要度最大的属性划入到正域中,重要度为零的属性划入到负域中,其余属性划入到边界域中;(3)循环计算边界域中属性的重要度,并将结果继续三分类直至约简结果满足约束条件。选取了8组UCI数据集,在传统属性约简和集成约简环境下分别进行实验。结果表明,在保证分类性能的前提下,新方法能够分别在2种环境下有效降低求解约简的时间消耗。Attribution reduction is one of the key topics in the field of rough set theory.Based on such theory,the concept of ensemble attribute reduction has been proposed.The ensemble reduction is to divide the sample into multiple decision systems in terms of the decision categories and then calculate them separately.Although ensemble attribute reduction balances the requirements of various decision classes,the corresponding time of attribute reduction is increased.To solve this problem,an attribute reduction acceleration method based on sequential three-way decisions is proposed.The specific steps are as follows:(1)The importance of the attribute in the decision system is calculated.(2)The attributes are divided into three groups in terms of the significance degree of corresponding attribute.Then,the attributes with maximal significance degree are classified into the positive domain,the attributes with zero significance degree are classified into the negative domain,and other attributes will be classified into the boundary domain.(3)The significance degree of the attributes in the boundary domain is calculated cyclically and the obtained result is divided,until the constraint is satisfied.8 UCI data sets are selected to conduct experiments in the traditional attribute reduction and ensemble reduction environments,respectively.The experimental results show that,under the premise of ensuring the classification performance,the proposed method can effectively reduce the time of attribute reduction in such two environments.

关 键 词：属性约简 邻域粗糙集 序贯决策 属性重要度 三支决策 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]