Research Of Fuzzy Clustering Algorithm And Cluster Validity Index

Clustering analysis as a method of data analysis applied very widely. Among them, the fuzzy clustering technology can deal with the uncertainty and fuzziness of things effectively,which can objectively reflect the real world and the nature of things has become an important part of data clustering theory. Fuzzy c-means(Fuzzy c- means, FCM)algorithm is one of the most commonly used fuzzy clustering technology, such as data mining, artificial intelligence and image processing. Although the FCM algorithm has the advantages of simpleness, high efficiency, easy to implementation, there are still some weaknesses and shortages, for example,it requires a predetermined number of clustering, clustering divided results are influenced by the choice of initial clustering centers, it is easy to fall into local optimal solution, it is sensitive to the noises and the isolated points, and so on.For the above some weaknesses existing in the FCM, we put forward the corresponding improve method, in this paper, the main works are as follows:(1) In the view of the problem that FCM algorithm automatically determines the clustering number of database, a novel validity index is proposed to determine the optimal number of clusters for fuzzy clustering. The novel validity index considers the degree of compactness, the degree of overlapping, and the degree of separation. The experimental results show that the optimal cluster number are obtained which are general used in FCM algorithm. the new index overcomes the shortcomings of FCM that the cluster number must be preassigned and works well in the situations when there are overlapping sub-cluster in the clusters.(2) In this paper, we combines FCM algorithm and PSO algorithm, redesigned the fitness function, and then proposed an improved clustering algorithm. The algorithm combines the FCM algorithm and PSO algorithm, because the advantage the PSO algorithm has stronger global searching ability and the fast convergence ability to solve the selection of the initial clustering center of FCM algorithm, and the new fitness function is designed by combining with the objective function by using the FCM algorithm and the clustering center from the two aspects. The experimental results show that to some extent the new improved algorithm can avoid the problem that FCM algorithm falls into local optimal problems easily.(3) At last, we combine the new index and the improved FCM algorithm, which applied into image segmentation. The algorithm uses CSO index to obtain the best image segmentation number firstly, combine the best image segmentation number with the improved FCM algorithm, and is applied to image segmentation lastly. The experimental results can achieve better segmentation effectively.