Research On Filling Incomplete Data Based On Fuzzification And Reduction Of Multi-label Decision Table

Fuzzy set theory and rough set theory are two types of important data mining tools.They can effectively deal with incomplete information systems and multi-label decision tables.This paper combines rough set theory and fuzzy set theory to discuss data filling of incomplete information systems problem and reduction problem of multi-label decision table.The specific researches are as follows:Based on the idea of fuzzy sets,this paper proposes fuzzy set-valued information systems based on traditional information systems.We obtain several binary relations in fuzzy set-valued information systems and study the relationships between them.Then,according to the different data types,this paper proposes two incomplete data filling algorithms:fuzzy set-value to fill incomplete data and set value(or real value)to fill in-complete data.In fact,these algorithms all need to transform the incomplete information system into a fuzzy set-valued information system.Experiments show that the incom-plete data filling algorithm proposed in this paper has higher accuracy and more stability than the traditional filling algorithm,and it is also useful for exploring the impact of incomplete data on the entire data set.Similar to the proposed method of fuzzy set-valued information systems,this pa-per induces fuzzy multi-label decision tables in multi-label decision tables.Next,in fuzzy multi-label decision tables,this paper gives different inductive methods of binary relations from three perspectives:micro,meso and macro.Different induce methods of binary rela-tions are proposed.Based on the fuzzy multi-label decision table,a new reduction meth od is proposed on the multi-label decision table,that is,the fuzzification-based reduction.The theory proves that compared with existing reduction methods,this reduction method can better retain the uncertainty degree of association between objects and label sets,and can degenerate to the existing reduction methods under certain conditions.Considering the application of multi-label data in practice,we also proposed the fuzzification-based specific label set reduction.Finally,this paper gives two heuristic algorithms to calculate a fuzzification-based reduction and a fuzzification-based specific label set reduction.