Research On Attribute Reduction Method Based On Dependence

Attribute reduction is one of the more and more attention in the field of data mining.Many scholars have done a lot of research on how to obtain reduction based on discernibility matrix,but because of the large storage space and low generation efficiency of discernibility matrix,this kind of attribute reduction algorithm is still not efficient.Therefore,in view of this problem,this paper has done the following three aspects of research:1)Based on the ability of distinguishing attributes,the concepts of comprehensive dependence degree and single attribute identification matrix are proposed.Because the elements of single attribute identification matrix are 0 and 1,the above problems are solved to a certain extent.Secondly,the nature and function characteristics of the comprehensive dependence are discussed,the progressive calculation method of the comprehensive dependence based on the single attribute identification matrix is given,and an attribute reduction algorithm based on the comprehensive dependence of dual attributes is designed.Finally,examples and experiments verify that compared with the reduction algorithm that adds an attribute successively,this algorithm is easier to obtain a reduction set with fewer attributes.2)On the basis of the study,in order to further improve the operating efficiency of the algorithm,this paper proposes the concept of local dependence and the sub matrix of the single attribute identification matrix,and gives a progressive expression based on the sub matrix of the local dependence calculation method.Furthermore,we designed an attribute reduction algorithm based on the local dependency,and verified the effectiveness of the algorithm through experiments.3)Since the classic rough set is not suitable for processing numerical data,researchers have proposed the concept of covering rough set.Therefore,in the context of covering rough sets,the ε-Boolean identification matrix is proposed,the calculation methods of dependence and local dependence are given.Finally,two attribute reduction algorithms are designed based on the degree of dependence and the degree of local dependence,and the effectiveness of this algorithm is verified through experiments.