Research On Fuzzy Clustering Algorithm And Its Validity

With the development of data mining, the research of clustering is becoming more and more deeper. At present there are many clustering methods and Fuzzy C-means clustering algorithm is the most widely used. To achieve better clustering result, many improved algorithms are proposed which are based on the algorithm. The effect of clustering needs the clustering validity index to evaluate. In the paper, the Fuzzy C-means clustering algorithm and cluster validity index has been improved from the following aspects:(1)Put forward the idea of attribute weights in the Fuzzy C-means clustering algorithm. The impact of the different attributes in the data set to the clustering is different, so it is necessary to assign different weights to different attributes. Based on the idea of attribute weights, this paper gets the new goal function, the cluster centers and the iteration formula of membership functions for Fuzzy C-means algorithm and by experiment to verify the improved clustering.(2) Put forward two new cluster validity indexes. As the diversity of clustering problem, there is no uniform cluster validity index to measure all results of clusteringes, so it is only improve from different cluster validity indexes. This paper put forward two cluster validity indexes, the first: based on the N. Zahid index, form an new cluster validity index throw improving the separation between class and class and the compactness within the class; the second: combinate the Vxie index, the Vbsaid index and the Vkwon index, introduce an new penalty factor and form an new cluster validity index. The two indexes are all verified by simulation that the stability, crucial and robustness are good.