Research On Cluster Algorithm Of Similarity Segmentation Based On Point Sorting

Cluster analysis is attempted to divide the whole data to some groups, in each one mostelements are similarity comparing with those in other groups, by setting a pair similarityelements together. As a data analysis technique, it have a large number of applications inmachine learning, pattern recognition, image analysis, data retrieval and bioinformatics andother fields. Because cluster analysis is to determine the closeness of the relationship betweensample based on attributes and characteristics of the sample by using mathematical methodsconsidering their degree of closeness to get a reasonable classification system. it necessary touse a computer cluster analysis method when there is a large amount of data which cannot beused for artificial classification to obtain significant results, or pre-treat data before analyzingthe data manually to obtain better classification results.Commonly used cluster analysis method can be divided into the following categories:classified clustering, hierarchical clustering, density clustering, grid clustering, clusteringmodel. These clustering methods are widely used in the previous studies, a category can beused as the representative by one or more of such analysis methods, such as k-meansclustering menthod can be a representative for classified clustering, DBSCAN as densitypolyethylene etc., and all kinds of methods have their advantages and disadvantages, so thateach methods should be applied to its suitable situation.This paper presents a segmentation clustering method based on similarity ranking points,in order to fully draw on the advantages of segmentation based on minimum spanning treeclustering algorithm in different situations. There are a variety of methods to treat theelements point to a similar clustering by sorting similarity between two elements, and dividethem in the points with large change degree of similarity to get clustering results. In order tocompare the performance of the method, the paper select the traditional clustering methodk-means, DBSCAN and newly popular method MCL algorithm to compare with the AP.Three sets of artificial classified data sets and four real classified data as test data, ourapproach obtained the excellent results in each set of test data by comparing selected methods.In practical problems, the size and dimension of elements which to be clustered are often amazing. It poses a severe challenge for the efficient of method of cluster analysis. Existingmethods cannot get better result when dealing with large-scale high-latitude data classresults.Therefore, the focus of future work will continue to conduct in-depth study of clusteranalysis and its improvement, so that it can better be applied to practical problems.