Improvement Of A Clustering Algorithm Based On Multicenter

Data mining is aim to find out connotative,unknown and potentially valuable knowledge and rules. Clustering analysis as one of the most important techniques in data mining, has received extensive attention in related fields. After years of development, cluster analysis technology has have made great development, is widely used in biology,information analysis, market analysis, pattern detection, character recognition, image detection, statistics and other fields.K-means algorithm is one of the most classic clustering algorithm, by virtue of simplicity, low time complexity has been widely studied and applied, however, due to using of single centre described a cluster, it cannot be effectively applied to data sets with arbitrary shape. Single cluster center can only describe convex-shaped cluster of defects, representative of the CURE,DBSCAN algorithm using multiple-point(or point) describes an arbitrarily shaped clusters. In recent years, academics have also raised the idea of arbitrary shaped segmentation, merging data sets, excellent results have been achieved.Based on the study of k-means algorithm and selection on the basis of the initial cluster centers, taking into account the larger or extended clusters in a data set into a number of globular clusters, and by methods that partition to be able to adapt to the data objects of arbitrary shape. Proposed an improved multi center clustering algorithm(IMCCA). The algorithm is first by a based on local density clustering center selection algorithm to select a set of cluster centers, these cluster centers have certain representative and uniform distribution in the high density region, as the center of the large clusters of arbitrary shape segmentation. Through between between the two clusters of connected vertex number description is segmented into small clusters of connectivity, the connectivity strength with small clusters. Experimental results show that based on local density clustering center selection algorithm to obtain the correct clustering center, and is not sensitive to the parameters, the algorithm IMCCA can effectively adapt to arbitrary shape, sizes, uneven density and with noise data set.