基于极大相容块的粗糙性度量及其属性约简

由于相似关系或相容关系不具有传递性或对称性,从而相容类或相似类之间存在误判,因此研究不完备信息系统中合适粒度下的粗糙性度量和属性约简算法很有必要。在不改变相关模型的基础上,文章通过极大相容块的思想,研究了非等价关系的基本知识粒度构造,进一步讨论了合适粒度下的粗糙性度量方法,提出了基于极大相容块的知识粗糙性更精确的定义和极大相容块的条件信息熵及其属性重要性定义,并证明了相关性质;给出了合适粒度下属性约简的启发式算法,结果表明,极大相容块的重要性度量避免了通常意义下粒度过粗问题,知识粗糙性更为准确。

Roughness measure and attribute reduction based on maximal consistent block

Due to the lack of transference and symmetry,there exists misjudgment in tolerance or similarity classes.Therefore,it is necessary to study roughness measure and attribute reduction algorithm with suitable granularity in incomplete information system.Without changing the relevant model,this paper studies the basic knowledge granulation of non-equivalence relation such as the similarity or tolerance classes,and the roughness measure method of knowledge with suitable granularity according to maximal consistent black(MCB).Based on it,the more accurate definition about the roughness of knowledge based on MCB,the conditional information entropy of MCB and its attribute significance are discussed.The relevant properties are proved and a heuristic algorithm of attribute reduction with suitable granularity is presented.The results show that the roughness measure based on MCB is more accurate,which can avoid the too rough granularity in the usual sense.