一种基于互信息增益率的新属性约简算法

为了获得决策系统中更好的相对属性约简，提出了一种基于互信息增益率的属性约简算法.该算法考虑了所选择条件属性与决策属性的互信息，还考虑了所选择属性的值的分布情况，从信息论角度定义了基于互信息增益率的属性重要性度量方法，并以此度量为启发式信息，算法从空集开始逐步将最重要的条件属性加入到选择属性集，直到所选择的条件属性集与决策属性集的互信息等于整个条件属性集与决策属性集的互信息时，算法停止.结果表明，算法能更有效地对决策系统进行约简，同时约简后的对象数目较少.

Novel algorithm for attribute reduction based on mutual-information gain ratio

To obtain good relative attributes reduction in decision systems, an algorithm for attributes reduction based on mutual information gain ratio was proposed. Both the value distribution of selected attributes and the mutual information between selected conditional attributes and decision attribute were considered. A new attribute importance measure method was defined from the viewpoint of information theory, and the measure was used as the heuristic information in the proposed algorithm. The most important condition attribute was added to the selected attributes set from empty set. The algorithm was terminated when the mutual information between the selected attributes set and the decision attribute set is equal to that between the whole condition attributes set and the decision attribute set. The experimental results show that the algorithm can effectively reduce the decision system, and that the number of objects after the reduction is small.