结合分辨矩阵改进的邻域粗糙集属性约简算法  

An improved neighborhood rough set attribute reduction algorithm combined with resolution matrix

作　　者：季雨瑄 叶军[1,2] 杨震宇 敖家欣 王磊 JI Yuxuan;YE Jun;YANG Zhenyu;AO Jiaxin;WANG Lei(School of Information Engineering,Nanchang Institute of Technology,Nanchang 330000,Jiangxi,China;Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing,Nanchang 330000,Jiangxi,China)

机构地区：[1]南昌工程学院信息工程学院,江西南昌330000 [2]江西省水信息协同感知与智能处理重点实验室,江西南昌330000

出　　处：《山东大学学报（工学版）》2022年第4期99-109,共11页Journal of Shandong University（Engineering Science）

基　　金：江西省教育厅科技项目(GJJ211920,GJJ170995);国家自然科学基金项目(61562061)。

摘　　要：针对基于邻域粗糙集属性重要度约简算法在某些决策表中约简正确率下降等问题,结合基于等价关系下的分辨矩阵知识,定义一种邻域决策系统下的分辨矩阵,邻域分辨矩阵由能够分辨不同邻域对象的条件属性子集组成。根据条件属性在邻域分辨矩阵中的占比提出一种属性重要度的度量方法,以新的重要度作为启发性因子,设计一种邻域决策系统下属性重要度启发性约简算法。该算法以核属性集作为初始集合,依次选择重要度大的属性加入到核集,直至找到最小属性约简时,算法终止。实例分析和UCI数据集试验结果表明,与基于属性依赖度的约简算法相比,该算法能够更有效地找到最小属性约简集,并且可以有效减少计算工作量,证明了算法的有效性和可实用性。Aiming at the problem that the reduction accuracy of attribute importance reduction algorithm based on neighborhood rough set decreases in some decision tables,combined with the knowledge of resolution matrix based on equivalence relationship,a resolution matrix under neighborhood decision system was defined.The neighborhood resolution matrix was composed of a subset of these conditional attributes that could distinguish objects in different neighborhoods.According to the proportion of conditional attributes in the neighborhood resolution matrix,a measurement method of attribute importance was proposed.Taking the new importance as the heuristic factor,a heuristic reduction algorithm of attribute importance in neighborhood decision system was designed.The algorithm took the core attribute set as the initial set,and successively selected the attributes with high importance to add to the core set until the minimum attribute reduction was found.The example analysis and the experimental results of UCI dataset showed that,compared with the reduction algorithm based on attribute dependence,the algorithm could find the minimum attribute reduction set more effectively and effectively reduce the computational workload,which could prove the effectiveness and practicability of the algorithm.

关 键 词：邻域决策系统 重要度 属性约简 邻域分辨矩阵 最小约简集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]