Rough算子求解不一致信息系统的约简算法

Rough Measure to Compute Reductions in Inconsistent Information Systems

为求解不一致信息系统的属性约简,在经典粗集理论模型的基础上,许多学者提出了上、下分布约简等方法,但是,这些方法尽可能保持了原决策系统的决策分布情况并且当数据集基数较大时,时间空间复杂度都较大。本文从另一个视角将大数据库中记录看成概率事件,利用粗集理论导出规则的模糊性度量方法—Rough算子,在多数优先的原则的基础上,将不一致信息系统转化为一致信息系统,并基于此提出了递增式反向求解方法。这种反向求解思想也为在大数据库中求解约简提供了可能。

In order to deal with inconsistent information system,there are many types of feature reduction such as upper/lower distribution reduction based on Pawlak rough sets theory.However,those methods reflect decision distribution of decision table,and the time/space complexity is often large when the cardinal number of feature attributes or records are large.In this paper,each object or row of a scale database table is viewed as probability cases and the inconsistent information can be changed into consistent information system using the majority precedence strategy.The fuzzy measure of cases is defined by the theory of rough sets,which named as rough measure.An incremental converse approach for computing reductions based on this definition is put forward.At first,the projection of single condition attribute is used to construct the 1-item denoted as L1 by the definition of condition attribute with respect to decision attribute.Accordiog to threshold,the successor of Lk is constructed.Gradually,the condition attributes in set of Lk is on way to the answer of reduction set.In addition,it is possible to get reduction in a large database because of converse approach.