基于多特征空间的粗糙数据分析方法  

Rough data analysis method based on multi-feature space

作　　者：杨习贝[1,2] 徐苏平[2] 戚湧[1] 於东军[3] 

机构地区：[1]南京理工大学经济管理学院,江苏南京210094 [2]江苏科技大学计算机科学与工程学院,江苏镇江212003 [3]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《江苏科技大学学报（自然科学版）》2016年第4期370-373,共4页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金资助项目(61572242;61272419;61305058;61373062);江苏省青蓝工程人才项目;中国博士后科学基金资助项目(2014M550293)

摘　　要：有关粗糙集的研究大多建立在原始属性空间之上,并未考虑具有不同决策类型的样本可能拥有自身特定的性质,而这些特性往往有助于产生更为精准的规则.文中针对决策系统中不同的决策类,为其构建了一系列新的能反映决策类自身性质的特征空间,在此基础上,给出了多特征空间下粗糙集的近似质量和条件熵的定义方法;最后通过实验分析发现构造的多特征空间就原始属性空间而言,决策系统的不确定性程度得以大幅降低,且分类性能亦有一定程度的提升.It is well known that most of researches about rough set are based on the original attribute space; they do not take into consideration extremely distinct characteristics which come from samples in different classes.These characteristics may contribute much to the accuracy of rules. To solve this problem,by different decision classes in the decision system,a series of new feature spaces are constructed which can reflect the characteristics of decision classes. Moreover,the definitions of approximation quality and conditional entropy in multi-feature space are also proposed. The experimental results show that,when compared with the original attribute space,the uncertainty of decision system is greatly reduced and the classification performance is also improved by using our multi-feature space.

关 键 词：多特征空间 模糊粗糙集 决策系统 条件熵 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]