Research On Complete Decision Rules Extraction Based On Upper Approximation Granule Description

With the deepening of social informatization and intelligence,intelligent decision-making problems in various fields have become particularly important,such as process optimization and fault diagnosis in the industrial field,intelligent diagnosis or prevention of diseases in the medical field.As a task of knowledge discovery,the extracted rules ’achieve the best balance between human and machine interpretability’,can provide an effective means for intelligent decision-making.Formal concept analysis is an effective granular computing method.Because its core data structure concept lattice has unique advantages in granulation and granular description of information,it has become a powerful tool for data analysis and rule extraction.The extracted decision rules from the decision formal context are mostly attribute implication rules,and the essence of rule extraction is generated by the granular description of the decision object granular subset.However,when the decision object granule is an indefinable granule,this granular description is often difficult to describe all the decision objects,resulting in incomplete extraction of decision rules,so that the extracted implication rules can only cover some parts of the decision objects.As an inevitable attribute analysis method,object-oriented concept lattices provide a research perspective different from common attribute analysis.Combined with the idea of upper and lower approximation,it can deal with inaccurate information well.Therefore,in order to extract complete decision rules,this paper studies the upper approximation granular description and complete decision rule extraction method based on necessary attribute analysis with the help of object-oriented concept lattice.The main research work is as follows:(1)When the decision object granule is an indefinable granule,the general granule description is often difficult to effectively describe all the decision objects,resulting in the loss of decision information.Therefore,an upper approximation granule description method based on necessary attribute analysis is proposed.Firstly,a type of (∨,∧)-definable granule based on necessary attribute analysis is defined.Secondly,it is proved that the definable granule based on common attribute analysis is also a definable granule based on necessary attribute analysis,and the upper approximate granule description method based on necessary attribute gives an upper approximate optimal description of the undefined granule,thus completely describing the decision object.Finally,the effectiveness of the method is illustrated by an example.(2)Based on the upper approximation granular description method proposed in(1),a complete decision rule extraction method is given.Firstly,the decision attribute is specified by the user to determine the decision object granule.Secondly,a ’But For’ complete decision rule is generated by the upper approximation granule and its description of the decision object granule.The concise description of the rule is obtained by the minimum generator.Then,the decision object granule is refined into the non-boundary decision object granule described by the non-boundary attribute of the upper approximation granule and the boundary decision object granule described by the boundary attribute,and the final complete decision rules are generated to provide more information for decision making.The effectiveness of the method is illustrated by an example.Finally,experiments are carried out on the early diabetes symptom data set.The results show that the complete decision rules extracted by this method can cover all the diabetic patients,thus providing a reference for self-monitoring of potential diabetic patients and high-risk groups.(3)A complete decision rule extraction prototype system for diabetes is implemented.Based on the theory proposed in(1)and(2),a prototype system for extracting complete decision rules for diabetes based on necessary attribute analysis is designed and implemented using python language.