决策表规则提取中分辨矩阵的降级算法研究

Research of algorithm for discernible matrix descending order in rule extraction of decision table

针对大型决策表规则提取中分辨矩阵级数高,计算复杂和低效的问题,提出了一种分辨矩阵的降级算法。算法以决策等价类为对象,构建简化分辨矩阵,其中每个元素都是由两个决策等价类构成的子分辨矩阵,与传统分辨矩阵相比,简化分辨矩阵的级数低、规模小,从而简化了属性约简,效率高;提取规则时根据需要灵活设置可信度,按照可信度值提取有效规则,使算法对不一致决策表也具有较好的适应性。实例计算表明算法清晰、简捷、有效。

Since discernible matrix is of higher order and rule extraction computing is complicated and inefficient in large decision table, an algorithm for discernible matrix descending order was proposed. Taking decision equivalence classes as objects, the algorithm builds brief discernible matrix. Every ele- ment of the brief discernible matrix is a subordinate discernible matrix constructed by two equivalence classes. The brief discernible matrix is of lower order and smaller scale than traditional discernible ma- trix. The attribute reduction is simplified and efficient. When extracting rules, according to the need, the algorithm can flexibly set reliability and select effective rules by reliability value. This can make the algo- rithm be of better adaptability to the inconsistent decision table. Example calculation shows that the algo- rithm is clear, simple and effective.