Generalization Improvement Of Rough Set Approximate Accuracy

Accuracy serves as a fundamental measure to quantify concept precision of rough sets,and it does not need to make assumptions about the value in advance.However,accuracy only considers the closeness of the upper and lower approximations,and does not consider the knowledge granularity information,so it has quantitative limitations.There are five existing types of approximate accuracy.They have been improved by introducing knowledge granularity,and they show different advantages.However,the existing research results lack a systematic analysis of them,and the related generalization mechanism has not been established.In this regard,this dissertation summarizes and analyzes five types of existing approximate accuracy,discusses the generalization principles of approximate accuracy based on knowledge granulation,and studies the construction of approximate accuracy models with parameters.In the five types of approxi-mate accuracy,one approximate accuracy with similarity is removed based on the generalization principles and the models with parameters.Then build specific improvement models for the re-maining four types of approximate accuracy.There are two main research contents.(1)In terms of example analyses,the granulation metrics and performance analysis of five types of approximate accuracy are compared,four types of improvement criteria(regarding mono-tonicity,consistency,irrelevance and lower-bound)are first established for approximate accuracy,and relevant satisfiability is inspected for five types of approximate accuracy.Then,two types of quantification optimizations(regarding approximation and monotonicity)are mined for approxi-mate accuracy,and concept accuracy and knowledge granularity are fused to set up two general-ized models with parameter.These two models unify the five types of approximate accuracy in form.(2)An approximate accuracy is removed based on the above-mentioned generalized para-metric models.The remaining four types of approximation accuracy are improved from two quantitative optimizations of“approximation and monotonic”respectively,and eight specific improvement models are proposed.It is revealed that they meet four types of improvement crite-ria and can be applied to two quantitative optimizations,and are effectively verified by knowledge examples and UCI data experiments.This dissertation focuses on the generalization improvement of rough set approximate accura-cy from the perspective of knowledge granulation.Through theoretical research,case analysis and experimental verification,the results obtained provide a generalized construction framework with approximate accuracy and specific improved models.These results are useful to the uncertainty characterization of rough sets and the application of knowledge quantification.