Research On Optimal Scale Selection Method Based On Multi-Scale Decision Table

Granular computing is a new concept and computing paradigm in the field of artificial intelligence.It is a study of thinking methods,problem solving methods,and information processing models based on multi-scale granular structures.According to different application requirements,choosing the most appropriate scale for data analysis has become an important research topic for granular computing.The existing optimal scale selection methods of multi-scale decision information tables have the following two restrictive characteristics:(1)for the same attribute,the optimal scale hierarchy of all objects is the same;(2)for the same object,under different attributes have the same optimal scale hierarchy.This paper proposes a new method for optimal scale selection based on attribute granularity tree.It studies the method of extracting rules of different scales that fuse different attributes,which is characterized by different scales of attributes in the same decision rule;For the same attribute,there are different scales under different decision rules,making the optimal scale more accurate and simplifying the rules.First of all,the attribute granular tree method in the formal context is introduced into the multi-scale decision information table.The concept of the attribute granular tree in the multi-scale decision information table is given,and its detailed properties are studied.In the granularity tree,the concept of cut that can describe the granularity hierarchy of attributes is introduced,and the partial order relation between different cuts is discussed.The concept of a mixed scale information table based on global cut was defined and its properties were studied.The judging conditions of optimal global cut for consistent and inconsistent complete multi-scale decision information tables are given,and the nature of optimal global cut is described using evidence theory,which is also implemented by algorithms.At the same time,the internal relations and differences between the existing optimal scale method and the optimal scale method in this paper are studied.Finally,a method of attribute reduction and rule extraction in a mixed scale decision information table based on optimal global cut is given,and an example is used to illustrate it.