Research On Granulation Data Clustering Method Based On Granular Computing Research

Granular computing is a vigorous research theme of information science and computer science. Its major aim is to solve problems on different granular structures. Problem-solving based on granular computing is multi-layer and multi-perspective. Its basic idea is to use information particles during the problem-solving process and describe, to infer and solve problems from different layers and perspectives. To a large extent,it can reflect intelligence in problem-solving. As an efficient granulating method, classical rough set method cannot handle the situation when the attribute values are continuous values. After discretization, the problem of information loss might occur. Therefore, this paper replaces the rough set with the fuzzy rough set for granulation. However, since the granulation method based on fuzzy rough set is supervised all through the process, it is necessary to consider a non-supervised granulation. The hyper-box granulation is a good granulation method, but the parameter selection might influence the granulation results. Concerning the problem, the hyper-box fuzzy clustering method is put forward to cluster data based on the hyper-box granulation results.Firstly,a granulation method of fuzzy rough set on non-uniform granular layers is put forward. Since traditional granulation methods mainly conduct granulation on a uniform granular layer, excessive roughness or fineness of the granular layer might result in problem-solving failures. This paper uses the attribute significance information to generate an incremental attribute subset sequence and induce a series of corresponding fuzzy relations. Based on the multiple fuzzy binary relations, the decision-type fuzzy rough set is considered in respective fuzzy binary relation. Since objects in the fuzzy lower approximation contain certain degree of similarity, this paper thinks that, when the membership degree of lower approximation is higher than certain threshold value, graining results can be generated. Information granular generated by every fuzzy binary relation is used to update the domain of discourse so as to induce a series of information granular. At last, the granular information is added to conduct data clustering and the clustering results can prove the efficiency of the granulation method proposed by in this paper.Secondly, since the parameter selection of the hyper-box can influence the granulation result, this paper puts forward a BFCM. A series of experiments are conducted and experimental results show that the new method can efficiently solve the problem of parametric influence.