不一致决策表的知识约简方法研究  

Approaches to Knowledge Reductions in Inconsistent Decision Tables

作　　者：李凡[1] 刘启和[1] 叶茂[1] 杨国纬[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都610054

出　　处：《控制与决策》2006年第8期857-862,共6页Control and Decision

基　　金：国家自然科学基金(天元)项目(A0324638)

摘　　要：目前计算不一致决策表的分布约简、最大分布约简和分配约简的方法均基于可辨识属性矩阵,在大数据集下耗时较多.为此,提出转化算法,将计算原不一致决策表的上述3种约简转化为计算3种一致决策表的Paw lak约简.通过应用针对后者的高效启发式算法,有效地减少了计算时间.此外,引入λ-约简的概念,通过调节λ的值,能得到一族反映决策矢量不同水平相似程度的知识约简.该方法降低了分布约简对决策表区分能力的过高要求,较上述3种约简更为灵活.Existing, approaches to knowledge reductions for the distribution reduct, the maximum distribution reduct and the assignment reduct of an inconsistent decision table are based on discernibility matrixes, which are very time-consuming when the dataset is large. To overcome this shortcoming, an approach is proposed to convert the computation for the three types of reducts of the original inconsistent decision table into the computation for the Pawlak reduct of three types of derived consistent decision tables. Thus, efficient heuristic knowledge reduction algorithms for the Pawlak reduct can be used to reduce computational costs. Furthermore, the λ-reduct, a new type of reducts, is introduced. By tuning the parameter λ , a set of reducts can be obtained, each of which reflects the different level of similarities of decision vectors. The λ-reduct eliminates the harsh requirements of the distribution reduct and is more flexible than the three types of reducts.

关 键 词：ROUGH集 知识约筒 不一致决策表 Fuzzy相似关系 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]