一种改进的基于差别矩阵的属性约简算法

为解决决策表属性约简完备算法约简质量低的问题，在基于差别矩阵的属性约简完备算法的基础上，引入信息论中信息熵和互信息增益的定义，给出一种启发式属性约简完备方法，通过实例说明启发式信息可以提高完备算法的约简质量，比较不同启发信息对完备算法的约筒质量和约简效率。试验结果表明，采用基于信息论定义的两种启发信息的完备算法约筒效率基本一致，该算法较非启发式完备算法有更好的约简质量。

Heuristic Algorithm for Attribute Reductions Based on Discernibility Matrix

Attribute reduction is one of the key problems of rough set theory. Quality issues always present in complete algorithm which is based on dicernibility matrix. Based on the complete algorithm, the definition of heuristic information based on information theory are introduced to effectively achieve the complete reduction of attribute. We have used two different definition of heuristic information in our algorithm: conditional information entropy and mutual information. The comparison of reduction quality and efficiency between heuristic algorithm and non-heuristic algorithm is examined and the experimental results show that this method was effective in attribute reduction and can obtain high quality reduction for most decision tables compared to non-heuristic algorithms.