The Domain Decreasing-Based Attribute Reduction Method Under Dominance Relations

With the progress and development of science and technology,tens of thousands of data are produced every day.How to dig out important and useful information from these data is particularly important.In practical problems,notice that some of the attribute values are preference-ordered,and attribute reduction as well as its efficiency have been the core problems of rough set theory.Therefore,the research of fast attribute reduction methods based on dominance relations is very meaningful.In attribute reduction methods of the dominance relation –based rough set model,it is necessary to calculate the upper and lower approximation sets as basic information granules,and the traditional algorithm needs to compare each attribute value of all the samples,consuming a large amount of time and memory.This will greatly affect the efficiency of computing the approximate sets and then reducing redundant attributes.Most of the existing acceleration algorithms are developed for computing equivalence classes,which can only be used to deal with symbol attributes.In this research,we propose a fast method to improve the computational efficiency of dominance classes by rapidly reducing the search space,so as to improve the performance of attribute reduction,especially when there are a large number of attributes.Furthermore,on the basis of quickly obtaining the dominant classes,the dominance classes and decision classes are then expressed as 01 matrices,and the positive region is calculated by fast “or” calculation.Finally,one attribute reduct is obtained by using the discernibility matrix method.The experimental results on UCI data s how that the method reduces the computation time and improves the efficiency of attribute reduction,while maintaining the same reduced attribute set.