An Improved Random Seed Searching Clustering Algorithm Based on Shared Nearest Neighbor

Abstract

Clustering analysis continually consider as a hot field in Data Mining. For different types data sets and application purposes, the relevant researchers concern on various aspect, such as the adaptability to fit density and shape, noise detection, outliers identification, cluster number determination, accuracy and optimization. Lots of related works focus on the Shared Nearest Neighbor measure method, due to its best and wide adaptability to deal with complex distribution data set. Based on Shared Nearest Neighbor, an improved algorithm is proposed in this paper, it mainly target on the problems solution of natural distribute density, arbitrary shape and cluster number determination. The new algorithm start with random selected seed, follow the direction of its nearest neighbors, search and find its neighbors which have the greatest similar features, form the local maximum cluster, dynamically adjust the data objects’ affiliation to realize the local optimization at the same time, and then end the clustering procedure until identify all the data objects. Experiments verify the new algorithm has the advanced ability to fit the problems such as different density, shape, noise, cluster number and so on, and can realize fast optimization searching.