Abstract: In practices, many real data in databases may vary dynamically. One has to run a knowledge acquisition method repeatedly in order to acquire new knowledge. This is very time-consuming. To overcome this deficiency, incremental approaches have been presented to deal with dynamic data set. This paper proposes a matrix-based incremental reduction approach with attribute values refining. When a part of data in a given data set is replaced by some new data, compared with the non-incremental reduction approach, the developed incremental reduction approach can find a new reduct in a much shorter time. Finally, experiments on two data sets downloaded from UCI show that the developed algorithm is effective.

Key words: attribute values refining, incremental learning, attribute reduction, rough set, knowledge granularity

摘要： 现实生活中许多数据库都是动态变化的，为了获取新的知识，传统的方法需要重复计算，耗时巨大。为了克服这个缺陷，有效处理动态数据，许多学者提出了增量学习方法。针对决策表属性值动态变化，提出了基于属性值细化的矩阵增量约简算法，当一部分属性值被细化时，同非增量约简方法相比，增量方法能快速找到新的约简，最后通过UCI数据进行性能测试，实验仿真结果表明所提增量约简算法是有效的。

关键词: 属性值细化, 增量学习, 属性约简, 粗糙集, 知识粒度