基于粗集理论的信息熵属性约简算法

Reduct Algorithm Based on Information Entropy and Rough Set Theory

本文针对粗集属性约简存在的问题,提出了一种基于信息熵的属性约简算法。算法中引入了信息熵的概念代替粗集约简g准则作为属性选择的标准,克服了粗集约简g准则对数据噪声的敏感性和不能表达属性间概率因果关系的缺点。本文通过两个实例表明,当属性间存在确定性关系时算法能够象粗集约简g准则一样找到表达这些关系的属性集;当属性间是概率因果关系,或确定性关系被数据噪声所掩盖,因而粗集约简g准则无法使用时,算法能够找到具有确定性关系的属性集,或是具有最小不确定性概率因果关系的属性集。

A new Information Entropy based reduct searching algorithm is proposed to tackle the problems involved in Rough set based reducting. Instead of Rough set reductingγcriterion, the new algorithm adopt information entropy as attribute selecting criterion. Using this algorithm, problems in prior criterion, such as sensitive to the data noise, unable to express the probability causality between attributes, can be solved. Two illustrated examples show that when there exists deterministic relationship between attributes, new algorithm can give the set of attributes expressing this relationship as the Rough set reductingγcriterion. When relationship between attributes is probability causative, or deterministic relationship is emerged by noise, Rough set reductingγcriterion becomes useless. However, in these cases, the proposed algorithm can give the correct answer needed.