Study On The Class Merging Cluster Algorithm Based On FCM

With the rapid development of database technology and widely used, more and more data is used. We hope that we can analyse it deeply, in order to make better use of these data. Supply capacity data and data analysis capabilities have become increasingly conspicuous, so it is the urgent need for automation technology, which is used to data processing deeply. Data Mining Technology comes into being. Cluster analysis in data mining technology is a classic content, an important tool for research.Fuzzy clustering can objectively reflect the real world because it can be described as a result of an intermediate category of samples. Cluster analysis has gradually become the mainstream. In a large number of fuzzy clustering algorithms, fuzzy C-means algorithm is most widely used and is the most sensitive. However, the algorithms are particularly sensitive to the initialization. It is easy to fall into local minimum or saddle point, and then not the global optimal solution. When we use this clustering algorithm, it is important to specify in advance the number of clustering data sets. However, the number of cluster C in general is very difficult to know in advance. For some irregular shape of the cluster, using Euclidean distance to describe the type of center is not appropriate; and FCM algorithm in general can only be found in clusters loaded ball.In this paper, we focuse on the research and analysis of FCM algorithm. We propose a new type of combined approach to improve the FCM algorithm. Clustering algorithm is divided into two stages.The first stage, using the combination of a minimum distance between a numerical algorithm for the Statute of technology to choose the initial cluster centers. The largest minimum distance algorithm can achieve the knowledge of input parameters to minimize. Numerical statute can greatly reduce the number of the samples and retain the distribution of samples.The second stage, merge the adjacent sub-category into big category using the physical meaning of FCM matrix. Complete the entire course of the final clustering.The main ideal: any extension of a large cluster or shape of the cluster can be expressed by a number of centers. Many central point is used to indicated the big category, and then merge these appropriate categories. In some extent, this method reduces the dependence on the initial cluster center and the number of the cluster. Clustering algorithm is not care whether the number of cluster is selected correctly; so we only need to provide a large enough number of initial clusters C.the final number of cluster is determined by merging the various categories. It is in line with the ideal of clustering.In order to verify the effectiveness and feasibility of the improved algorithm, we carry out a number of experiments.It is verified that the quality of the clustering and cluster stability are much better than FCM algorithm.In this paper ,the improvement of FCM is feasible and effective.