Abstract: Interval-valued decision information systems are generalized models of single-valued information systems. A kind of [α] maximal tolerance class is introduced by similarity grade of attribute’s interval-value in interval-valued decision system. This paper defines new conditional entropy among attributes in interval-valued information systems and proposes two types of measurement of relative attribute importance, which is inner attribute importance and outer attribute importance. Furthermore, a heuristic relative attribute reduction algorithm based on[α]conditional information entropy is given, and the validity of the algorithm is illustrated by some experiments.

Key words: interval-valued decision system, [α]maximal consistent class, [α]approximation reduction, relative attribute importance

摘要： 区间值决策信息系统是单值信息系统的一种推广，借助于属性区间值的相似程度在区间值决策系统上引入[α]极大相容类的概念，定义了一种新的条件信息熵，提出了相对属性内（外）重要度的度量方法，进一步，给出基于[α]条件信息熵的启发式相对约简算法，通过实验验证了该算法的有效性。

关键词: 区间值决策系统, [&alpha, ]极大相容类, [&alpha, ]近似约简, 相对属性重要度