Research On Multi-Granularity Attribute Reduction Based On Compatible Rough Set

Attribute reduction is an important research direction in the field of data mining,which aims to get rid of redundant conditional attributes and get a more concise description of the original data set.However,the attribute reduction of classical rough set theory uses the equivalence relation to partition the particles,and uses the positive region to measure the attribute importance,the description of concepts is not fault-tolerant,.In addition,it is necessary to discretize the data before attribute reduction,and the discretization will result in the loss of attribute information.At the same time,the influence of boundary region and its change on the result of decision-making is neglected when the attribute importance degree is characterized by positive region,which will affect the result of attribute reduction.In order to solve these problems,combined with the analysis of the difference between symbolic and numerical attributes,the feature support compatibility relation is defined,which can describe the similarity between objects in the mixed attribute data set.on this basis,a consistent rough set model based on feature support is constructed,and the attribute reduction of multi-granularity spatial mixed data under this model is studied.The main research contents are as follows:(1)Based on the consistent relationship of feature support,a consistent rough set model of feature support is constructed.Finally,combined with the idea of multigranularity problem analysis,a multi-granularity attribute reduction algorithm based on feature support degree(MGARCRSFS)is designed.The experimental results of UCI data sets show that MGARCRSFS is an efficient method for attribute reduction of hybrid data sets.(2)Aiming at the problem of the influence of the uncertainty from boundary domain on decision classification,a new definition of boundary is improved in this paper to simplify the calculation process of positive domain,a multi-granularity attribute reduction method based on consistent rough sets with double boundaries(MGARCRSDB)is proposed.The experimental results of UCI data set show that MGARCRSDB is better than forward attribute reduction algorithm.