Research On Uncertainty Measurement And Attribute Reduction For Entire-Ggranulation Rough Sets

With the arrival of big data's era,the amount of data increases rapidly.All works of life are facing massive data information,and more companies are "drowning" in massive data.how to excavate effective information has become a hot topic.Rough set theory can effectively deal with data and information in complex systems,and become mathematical tool for dealing with imprecise and uncertain problems.It's greatest advantage is that process the data without any prior knowledge and find the best information.Not only keep the ability of knowledge classification unchange,but also can redundantly process the data(attribute reduction),and finally obtain the effective information.As a new type of rough set model,entire-granulation rough sets are a dynamic and static combining rough set model.It has a certain quantum computing idea,which can partly express complexity,diversity and uncertainty of human cognition.Compared with other traditional rough set models,entire-granulation rough set is dynamic and can jump from one granularity to another,jump naturally and without any difficulty.Compared with multi-granularity rough set,it is a superset of multi-granularity rough set,which is more comprehensive than multi-granularity rough set.Two core the research of rough set theory is Uncertainty analysis and attribute reduction.Because of it's dynamic change,entire-granulation rough sets are more flexible and diverse in dealing with uncertainty,and the study of entire-granulation attribute reduction is also rich.With the development of the theory of entire-granulation rough sets,the study of uncertainty and attribute reduction arethe important tasks at present.In this paper,the uncertainty measurement and attribute reduction of entire-granulation rough set are systematically studied.The specific work are as follows:1.A new uncertainty index of entire-granulation is defined.Based on the traditional rough set theory and the characteristics of entire-granulation rough set model,uncertainty index of entire-granulation is defined,which is easy to understand the uncertainty in the entire-granulation rough set theory,which enrich connotation of entire-granulation rough set,laying a theoretical foundation of entire-granulation attribute reduction.2.Redefine related entire-granulation attribute reduction.Attribute reduction of entire-granulation for a single concept and entire-granulation Pawlak reduction are redefined by using uncertainty index of entire-granulation proposed in this paper,which not only simplifies the complexity of the original definition,furthermore,The obtained heuristic algorithms of entire-granulation attribute reduction.3.Two types for attribute reduction are proposed.In this paper,the concepts of distinguishability and positive-region-based-distinguishability are defined from the point of view of knowledge division,and the corresponding attribute reduction criteria are proposed,explore the relationship between these two attribute reduction and all kinds of entire granulation attribute reduction.Theoretical analysis shows that attribute reduction based on positive region distinguishability and entire-granulation Pawlak reduction are completely equal,entire-granulation Pawlak reduction can be obtained indirectly by means of positive region distinguishability attribute reduction.Then a heuristic algorithm based on positive region distinguishability attribute reduction is designed.4.The efficiency and feasibility of the algorithm based on positive region distinguishability attribute reduction are analyzed experimentally.In this experiment,the method of comparative analysis and cross-validation is used to compare the algorithm with other methods,and attribute reduction is carried out in six sets of data sets respectively.Finally,the reduction time,reduction rate and classification accuracy are compared.