一种基于二进制分辨矩阵的属性约简新算法  

New algorithm for attribute reduction based on binary discernibility matrix

作　　者：赵军[1] 陈宸[1] 

机构地区：[1]重庆邮电大学计算机科学与技术研究所,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2012年第4期490-494,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市自然科学基金(2008BA2017);重庆市教委科学技术研究基金(KJ080510;KJ110522);重庆市杰出青年科学基金(2008BA2041)~~

摘　　要：决策表的属性约简是粗糙集理论中的关键问题之一,高效启发式约简算法有待深入研究。基于二进制分辨矩阵,定义"加权重要度"来度量属性相对重要性;这一概念综合利用分辨矩阵行和列2个方向的特征,将其集成为一个归一化的概念,因而在排序过程中能有效避免从行、列2个方向独立排序。基于"加权重要度"概念,提出一种新的启发式属性约简算法;仿真实验结果表明,与参考算法相比,新算法有效缩短了运行时间,更有可能获得系统的最优约简。The attribute reduction of decision tables is one of the key issues in rough set theory. Effective and efficient heu- ristic reduction algorithms are deserved to be further studied. Herein, a concept of ＂weighted importance＂ is defined based on binary discernibility matrix to heuristically measure the signifieances of attributes of a decision table. The concept com- prehensively makes good use of the characteristics of columns and rows of binary discernibility matrix and integrates them into a normalized concept ; consequently, independent sorting computation in respective directions of column and row is effec- tively avoided. Then a new heuristic algorithm for attribute reduction is proposed based on concept of ＂weighted impor- tance＂. Simulation experiment resuhs show that compared with some benchmark algorithms, the new algorithm runs faster and is more possible to get optimal reductions of decision tables.

关 键 词：粗糙集理论 决策表 属性约简 二进制分辨矩阵 加权重要度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]