基于条件熵的不完备信息系统属性约简算法

在相容关系下定义了三种不完备条件熵——H′条件熵、E′条件熵和I′条件熵,并对它们的性质进行了分析比较,研究发现,H′条件熵和I′条件熵不适用于相容关系下信息观点的约简。利用E′条件熵刻画信息系统中属性的相对重要性,设计了一种新的基于信息论观点的启发式约简算法,它统一了完备信息系统与非完备信息系统中的约简方法。通过实例说明,该算法能得到决策表的相对约简。

Attribute Reduction Algorithm Based on Conditional Entropy under Incomplete Information System

Knowledge reduction is an important issue in data mining. This paper focuses on the problem of attribute reduction in incomplete decision tables. Three types of incomplete conditional entropy are introduced based on tolerance relation, such as H′ conditional entropy, E′ conditional entropy, and I′ conditional entropy, which are proved to be an extension of the concept of conditional entropy in incomplete decision tables. Compared with H′ and I′ conditional entropy, E′ conditional entropy decreases monotonously with the amount of attributes. Based on E′ conditional entropy, a new reduced definition is presented, which integrates the complete and incomplete information systems into the corresponding reduced algorithm. Finally, the experimental result shows that this algorithm can find the reduct of decision tables.