摘要： 提出了一种处理海量的不完备决策表的方法。将基于互信息的属性重要度作为启发式信息，利用遗传算法对不完备的原始决策表中的条件属性进行约简，形成包含missing值的决策表，称为优化决策表。利用原始决策表自身的信息，通过属性扩展，从优化决策表中抽取一致性决策规则，而无须计算missing值。该方法在UCI的8个数据集上的实验结果优于EMAV方法，是一种有效的从海量不完备决策表中抽取规则的方法。

关键词: 粗糙集, 不完备决策表, 互信息, 不规则决策规则

Abstract: By regarding the significance of attributes defined from the viewpoint of information theory as heuristic information, and introducing the heuristic information into genetic algorithm which takes attribute lengths as fitness function, this paper proposes an effective heuristic genetic algorithm for minimizing relative reduction. It uses the heuristic genetic algorithm to select the important attributes from the original decision table, and forms an optimal decision table which contains minimal missing value. A method for extracting rules in optimal decision table based on attribute extension is presented. It can get the rule sets without estimating missing attribute value. Experiments justify that the accuracy of obtained rule set is better than the highest accuracy among those of the estimating missing attribute value methods.

Key words: rough set, incomplete decision table, mutual information, irregular decision rule

中图分类号: 

• TP301.6