Abstract: Attribute reduction is one of the most important research issues in the rough set theory. In recent years, the attribute reduction problem under the rough set theory has attracted many scholars?? attention. However, most of the attribute reduction methods are proposed based on the indiscernibility relations or the discernibility relations, and the performance of the attribute reduction only depends on the changes of the equal classes or the approximation sets, but the changes between different clusters of objects that do not have an equivalence relationship are ignored. Therefore, this paper proposes a notion of the inter-class separability. Compared with the equivalence class or the approximation sets, it can reflect the degree of distinction between different classes as the attributes change. This paper explains and defines the inter-class separability and the inter-class coincidence degree, and proposes an inter-class separability based attribute reduction method by considering the heuristic search strategy. Moreover, the experimental results indicate the efficiency of the proposed attribute reduction method.

Key words: attribute reduction, rough set theory, inter-class coincidence degree, inter-class separability

摘要： 属性约简是粗糙集理论中最重要的研究问题之一。近年来，粗糙集理论下的属性约简问题引发了学者们广泛的关注。然而，大多数属性约简方法都是基于不可分辨或可分辨关系所提出的，属性约简的性能仅仅取决于等价类或近似集的变化，却忽略了不具有等价关系的对象所在的不同类簇间关系的变化情况。因此，引入了类间区分度的概念，相较于等价类和上下近似集而言，它可以反映类簇区分程度随属性变化而变化的情况。对类间重合度和类间区分度进行了解释及定义，并结合启发式搜索策略，提出了一种基于类间区分度的属性约简方法，实验验证了所提方法的有效性。

关键词: 属性约简, 粗糙集理论, 类间重合度, 类间区分度