南京大学学报（自然科学版）Issue(4)：714-722,9.DOI:10.13232/j.cnki.jnju.2015.04.009
粗糙集的不确定性度量比较研究
Comparative study of uncertainty measure in rough set
摘要
Abstract
Rough set theory is one of important soft computing tools,which has become immediate areas of research focus in data mining and granular computing,and extensively being employed in many practice application,such as fault diagnosis, fraud detection and so on.Uncertainty measure is one of the critical issues in rough set theory,which can measure the dependence and similarity between two attribute sets and be used to evaluate the significance of attributes in attribute reduction algorithms and clustering algorithms.The uncertainty measures,whose capabilities of characterizing the dependence and similarity between attributes are significantly different,largely affect the outcomes of attribute reduction al-gorithms and clustering algorithms.To detect the difference of them,we first analyze the differences among the common de-pendence measures,such as accuracy of approximation classification,quality of approximation classification,Shannon condition entropy and complete condition entropy.And we find out that there exists an order of capability of assessing de-pendence,i.e.complement entropy is superior to Shannon entropy,and Shannon entropy is superior to accuracy of approximation classification and quality of approximation classification.Furthermore,we analyze the difference between the similarity measures,such as Shannon mutual information and complement mutual information,and obtain the sequences of similarity degrees,i.e.Shannon mutual information is superior to complement mutual information when they are used to assess the similarity between two attributes.Thus it is clear that our conclusion,from theoretical aspect,can help to guide how to choose an appropriate uncertainty measure in the process of attribute reduction and clustering analysis.关键词
粗糙集/不确定性度量/条件熵/互信息Key words
rough set/uncertainty measure/condition entropy/mutual information分类
信息技术与安全科学引用本文复制引用
魏巍,魏琪,王锋..粗糙集的不确定性度量比较研究[J].南京大学学报（自然科学版）,2015,(4):714-722,9.基金项目
国家自然科学基金(61303008,61402272),山西省高等学校科技创新项目(2013102),山西省自然科学基金(2013021018-1) （61303008,61402272）
