Abstract: To solve the problem of choosing a best reduction from several reductions of one decision table, after thinking about the average significance of attributes in reduction and the number of attributes, a new definition named attribute significance was proposed and proved in detail. Then the algorithm of finding the best reduction was presented based on this new definition. Finally, examples demonstrate the usefulness of the algorithm.

Key words: rough set, attribute reduction, attribute significance, relative indispensable attribute, artificial neural network

摘要： 为解决多约简决策表的约简选取问题，在综合考虑约简中属性的平均重要性以及属性个数的基础上,提出了约简重要性的概念,并对其进行了详细论证。以此概念为基础介绍了最佳约简求解算法，即以此概念为准则对多个约简进行比较，可选出一个最佳约简。最后以实例验证了算法的实用性。

关键词: 粗糙集, 属性约简, 属性重要性, 相对必要属性, 人工神经网络