改进的差别矩阵启发式属性约简算法

Improved difference matrix heuristic attribute reduction algorithm

为在决策表中获得更好的属性约简组合,从信息论角度分析,在基于区分矩阵的基础上,提出一种改进的以条件熵作启发信息的约简算法。同时考虑条件属性相对于决策属性的条件信息熵以及属性值的分布情况,用它们的比作为启发因子,重新给出一种度量属性重要度的依据,得到属性约简集。实验结果表明,该算法能够有效约简属性集,使约简结果获得最简决策规则组合。

To get a better combination of attribute reduction in decision table,from the viewpoint of information theory,the improved algorithm was proposed,in which the conditional entropy was taken as heuristic information for attribute reduction based on the discernibility matrix.The conditional entropy of condition properties with respect to the decision attribute and its value distribution were considered at the same time.An attribute importance measurement method was presented by using their ratio as heuristic factor,and eventually getting attribute reduction sets.Experimental results show that the attributes sets of decision table can be effectively reduced using the proposed algorithm,and the reduction results can obtain the smallest combination of decision rules.