基于粗糙集的区间型数据离散化算法  

Rough sets based discretization algorithm for interval data processing

作　　者：谭旭[1] 唐云岚[1] 陈英武[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,长沙410073

出　　处：《系统工程理论与实践》2009年第6期157-165,共9页Systems Engineering-Theory & Practice

摘　　要：针对条件属性取值为区间型数据的离散化问题,提出了一种新的基于粗糙集理论的离散化算法.首先将粗糙集理论中上、下近似的概念进行扩展,用以描述区间数对象间的距离和相似关系,并通过定义相似度阈值来确定对象间的相似关系.为了达到用最少的离散划分区间得到较好的离散化结果,并合理地确定相似度阈值,文章给出了粗糙熵的概念.通过离散化属性的上、下近似粗糙熵值的计算以及该属性下各区间数对象的相似度矩阵的确定,可以得到该属性下最终的离散化结果.最后给出了一个烟叶质量等级评价的实例,实验结果表明该算法是有效的.Aiming at discretizating condition attributes with interval value, we proposed a new discretization algorithm based on rough set theory. First, we extend the upper and lower approximation concept from rough set theory, so we can describe the distances and similarity relations among objects in an interval. Then similarity relations among the objects are determined by defining similarity threshold. In order to use minimal discretization intervals while obtaining optimal discretization results and define reasonable similarity threshold, we propose a new concept： rough entropy. By calculating upper and lower approxi- mate rough entropy and determing similarity matrix of these objects, we can obtain the final discretization result. At the end of this paper, a tobacco leaf quality assessment example is given to prove the feasibility of this algorithm.

关 键 词：粗糙集 区间型数据 离散化 相似度矩阵 相似度阈值 粗糙熵 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]