一种基于属性度量的快速属性约简算法  

A Quick Algorithm for Attribute Reduction Based on Attribute Measure

作　　者：钱文彬[1] 杨炳儒[2] 谢永红[2] 李慧[2] 

机构地区：[1]江西农业大学软件学院,南昌330045 [2]北京科技大学计算机与通信工程学院,北京100083

出　　处：《小型微型计算机系统》2014年第6期1407-1411,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61175048)资助;科技部创新方法专项项目(2010IM020900)资助;材料领域知识工程北京市重点实验室2012年度阶梯计划项目(Z121101002812005)资助

摘　　要：针对现有基于属性度量的属性约简算法未考虑决策表中各属性信息粒度的差异性和计算复杂度不理想的情况,首先给出了决策表信息粒度的定义,同时从理论上分析了核属性和粒度细化在属性约简中的重要性,由此构造了一种新的混合属性度量方法,使得求解的属性约简更符合实际情况,并能有效缩小了算法的搜索空间,在此基础上,设计了一种基于属性度量的快速属性约简算法,算法能求解出决策表的完备属性约简;最后,通过实例分析和实验结果进一步验证了算法的可行性和有效性.In view of existing attribute reduction algorithms based on attribute measure exists two main shortcomings： the differences of information granular of each attribute are not considered in decision table, and some algorithms are usually computationally timeconsuming for large data. To overcome these shortcomings, firstly, the definition of information granularity of decision table is introduced. What＇s more, the importance of core attributes and granularity is analyzed, so a new attribute measure method is constructed, it makes attribute reduction is more consistent with the actual situation, and can reduces the searching space of attribute reduction algorithm, on this basis, a quick attribute reduction algorithm based on the proposed measure is presented, the algorithm can calculate complete attribute reduct; Finally, case analysis and experimental result further verify the feasibility and efficiency of the algorithm.

关 键 词：决策表 属性度量 信息粒度 属性约简 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]