基于邻域组合熵的属性约简算法  

ATTRIBUTE REDUCTION ALGORITHM BASED ON NEIGHBORHOOD COMBINATORIAL ENTROPY

作　　者：王光琼 Wang Guangqiong(School of Intelligent Manufacturing,Sichuan University of Arts and Science,Dazhou 635006,Sichuang,China;Dazhou Industrial Technology Institute of Intelligent Manufacturing,Dazhou 635006,Sichuan,China)

机构地区：[1]四川文理学院智能制造学院,四川达州635006 [2]达州智能制造产业技术研究院,四川达州635006

出　　处：《计算机应用与软件》2018年第12期269-273,284,共6页Computer Applications and Software

基　　金：省教育厅重点项目(18ZA0419;18ZA0421)

摘　　要：为了对符号性和数值型属性共存的邻域信息系统进行属性约简,从信息论的视角出发,定义邻域信息系统中的邻域条件熵。同时考虑知识不确定性和集合不确定性对属性重要度度量的影响,结合邻域条件熵和邻域近似精度,定义一种新的属性重要度度量——邻域组合熵。给出邻域组合熵的相关定理,提出基于邻域组合熵的属性约简算法。在UCI数据集上的实验表明,该算法能够获得约简集较小而分类精度较高的约简结果。In order to reduce the attribute of the neighborhood information system with the coexistence of symbolic and numeric attributes, the neighborhood conditional entropy in the neighborhood information system was defined from the perspective of information theory. We also considered the influence of knowledge uncertainty and set uncertainty on the measure of attribute importance. Combining the neighborhood conditional entropy with neighborhood approximation accuracy, we defined a new attribute importance measure, neighborhood combination entropy. The correlation theorem of neighborhood combinatorial entropy was given, and an attribute reduction algorithm was proposed based on neighborhood combinatorial entropy. The experiment on UCI data set shows that the algorithm can get the reduction result with smaller reduction set and higher classification accuracy.

关 键 词：属性约简 组合熵 邻域信息系统 不确定性 属性重要度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]