Research On Three-way Decision Model For Boundary Processing Based On Category Feature Representation

In the “explosive” era of big data,the data generated in real life often has the characteristics of low quality,especially in the process of data decision-making,there will be many uncertainties.Therefore,in the context of big data and artificial intelligence,how to mine effective information for correctly handling these uncertain data is one of the important research directions of current data researchers.For this problem,there have been many methods and theories to deal with related problems,three-way decision theory has made a big important contribution,and has always been one of the popular research methods to deal with uncertainty.The core idea of the three decisions is to divide the positive and negative domains corresponding to the certain samples,and at the same time temporarily divide the uncertain samples into the boundary domain.Aiming at the situation that the traditional two-way decision theory only "accept" and "reject" in the data classification process,the three-way decision adds a third kind of decision choice to uncertain data,and the classification decision result no longer only has "accept" and "reject" " two choices,but delayed decision on this part of the data.That is,when the data information is not enough to support the acceptance decision or the rejection decision,it is temporarily divided into the boundary region,and the delayed decision is taken,and further decision processing will be made when more suitable information is subsequently tapped.Therefore,in the data decision-making process,the three-way decision theory can effectively deal with clear and uncertain samples,especially in processing uncertain data in boundary region.However,further research is needed on how to use certain data samples to guide the solution of uncertain data.In order to better deal with the decision problem of data,this paper starts from a clear data perspective and mines useful information to guide the solution of uncertain data.Combining the theory of three-way decision theory with the minimum cover algorithm(MinCA)to form a three-way decision model.Among them,the biggest advantage of the MinCA algorithm is that it does not require any parameters,the distance between the data sample and the coverage center is directly compared with the size of the coverage radius to determine the category of the data sample,forming three regions: Positive region,Negative region,Boundary region(POS,NEG,BND).Then,based on fuzzy quotient space theory,establish fuzzy equivalence relations in POS and NEG,respectively.And it obtains hierarchical feature representations of category region,and chooses the most suitable combination of feature representations to deal with the boundary domain with uncertain data,which can improve the overall classification accuracy of the data.The main content of this paper includes the following three points:(1)The paper first elaborated on the classification of uncertain data,the study of three decisions in data classification and the study of boundary domain processing,focusing on the analysis of the role and advantages of the three-way decision in boundary processing.Then,based on the three-branch decision theory,the three-way decision classification model based on the minimum cover algorithm is introduced in detail,and the division process of the three regions is shown.The fuzzy quotient space theory is introduced,starting from the basic definition and principle of the quotient space theory.It highlights how to establish a fuzzy equivalence relationship,which lays a solid theoretical foundation for the establishment of the feature representation part of this paper.Finally,for the classification of data,this article selects Accuracy as the evaluation index,and gives a detailed introduction and explanation.(2)For solving the boundary problem,which lack sufficient information,we propose a hierarchical feature representation for boundary processing based on three-way decision model(HFR-TWD).First,it combines the minimum cover algorithm with three-way decision theory,which divide the data into POS region and NEG region with clear information and BND region with uncertain information.Then,based on fuzzy quotient space theory(FQST)we construct two kinds of fuzzy equivalence relation matrix from POS region and NEG region,respectively,and obtain the corresponding hierarchical feature representation through the processing of the definition of the Cut relation.Finally,the feature representation of the high-precision layer is obtained through the verification of the boundary domain sample,further processing uncertain boundary domain samples to form the final binary decision result.Experimental results show that the algorithm can effectively process uncertain boundary samples.Compared with other comparison algorithms,it can effectively improve the accuracy of data classification.(3)From the perspective of granularity analysis,for the high-precision layer feature representation selected during the HFR-TWD algorithm,it is found that there exists big difference between the high-precision layer and the upper/lower layer feature representation,there is no continuity between them,and that is not the optimal selection.Therefore,this paper proposes an Adaptive hierarchical feature representation model based on three-way decision theory(AH3)with problem oriented for boundary processing.First,based on MinCA model,all clear samples are divided into POS region or NEG region,and the uncertain samples are divided into BND region.Then,based on the fuzzy quotient space theory(FQST),separately process the POS and NEG region samples to construct a fuzzy equivalence relationship,in the process,in order to strengthen the relationship between the features and other important features,at the same time,delete weak features,we introduced the function of the variance,and Combine it with the mutual information to obtain a hierarchical feature representation with higher correlation.Then,through boundary domain verification,obtain the feature representation layer with the high-precision in POS region and NEG region.Finally,from the perspective of granularity,it adaptively decomposes the feature representation with the high-precision layer between the upper layer and the lower layer to finer granularity,two adaptive granular spaces will be selected from two regions,and two adaptive layers are properly for boundary processing.Experimental results show that the feature representation obtained by the algorithm can process the boundary domain samples more effectively,and further improve the classification accuracy of the overall data.