Research On Uncertainty Measurement Method For Neighborhood Rough Set Model

Rough set theory,as a mathematical tool that can deal with inaccurate,inconsistent and incomplete data,has received wide attention from experts at home and abroad for its remarkable performance in uncertainty measurement and its characteristic that no prior knowledge is required in the processing.Numerical data is a common type of information system at present.Due to the insufficiency of classical rough set model in processing numerical data,scholars proposed models such as neighborhood rough set model to better handle numerical data.Neighborhood rough set model being an important model,the uncertainty measurement of which is also the focus of the research.Aiming at the uncertainty measurement method of neighborhood rough set model,I mainly did the following research:(1)The max-decision neighborhood rough set model is an effective model,but it lacks an effective measure.In this paper,the uncertainty measure of the max-decision neighborhood rough set model was discussed.The measures of max-decision neighborhood precision,maxdecision neighborhood roughness and max-decision neighborhood granularity under this model were first given respectively,and it was proved that the proposed measurement methods have good uncertainty measurement effect.In order to describe the uncertainty of the neighborhood rough set model more comprehensively,the definition of mixed rough granularity was given at last.This measure,combining the advantages of max-decision neighborhood roughness and max-decision neighborhood granularity,is a more comprehensive measure.Finally,an experiment was carried out on the UCI open data set and the results showed that the uncertainty measurement method proposed in this paper had a better measurement effect and the proposed algorithm had a higher accuracy in the classification results after attribute reduction.(2)There are two problems in the existing uncertainty measurement methods in the neighborhood rough set model: One is that the uncertainty measurement method based on boundary region does not distinguish the influence of objects belonging to the objective concept and objects not belonging to the objective concept.The other is that soly relying on the pure information entropy and conditional entropy will amplify the uncertainty of the rough set to a certain extent,this paper improved the adding the change of boundary region into the design of uncertainty measures,providing the definition of boundary region information entropy and boundary region condition information entropy,and verifying related properties.This article combined roughness,boundary region information entropy and boundary region condition information entropy,provided the definition of the rough boundary information entropy and rough boundary condition entropy.These two measures combining the algebra view of neighborhood rough set and information view of uncertainty measurement is a more comprehensive measure of uncertainty.Finally,an experiment was carried out on the UCI open data set.The experimental results showed that the four uncertainty measures proposed in this paper had good measurement effect and in the attribute reduction and classification experiments,the rough boundary information entropy and the rough boundary condition entropy had better reduction effect and higher classification accuracy.