Optimal Granularity Selection In Different Multi-granular Labeled Decision Systems

In the practical application,the multi-granular labeled decision system is often used in various information analysis,the key to acquire knowledge is how to quickly select the most needed granularity from the system.Both rule extraction and attribute reduction are difficult before the optimal granularity is selected.Therefore,optimal granularity selection plays an important role in knowledge acquisition of the system.In this paper,the multi-granular labeled system with multi-decision and the multi-granular labeled decision system with multi-source is defined.Then,the ways to choose the optimal granularity for the new systems is discussed,and how to update the optimal granularity when the attribute value changes is studied.In this paper,the main parts of the new research are as follows:(1)The multi-granular labeled system with multi-decision is defined,and the distribution function,maximum distribution function,generalized decision function and lower approximate function applicable to the system are given.Then,the optimal granularity is discussed by considering the granularity maintaining distribution consistent,maximum distribution consistent,generalized decision consistent or lower approximation consistent.Finally,the corresponding examples are given to prove the feasibility of the method.(2)The multi-granular labeled decision system with multi-source is defined,and the strategy of selecting the optimal granularity in the system is discussed.Specific strategy:firstly,the optimal granularity of each information source are selected,and then the optimal granularity of each information source is fused to select the global optimal granularity.Three methods are given according to different application requirements,there are "the coarsest","the finest" and "the vast majority",respectively.At the same time,corresponding examples are given to prove the feasibility of the method.(3)How to update optimal granularity in multi-granular labeled decision systems under the environment of attribute value changing is researched.The corresponding static and dynamic optimal granularity selection algorithm are designed,respectively.Finally,six UCIdata sets are selected for experiments.It is worth pointing out that the experimental results prove that the new method is feasible and effective,and the results are the same.