A novel hybrid structure for clustering

Abstract

Conventional clustering methods such as k-means or subtractive clustering need some information include the optimal number of clusters, otherwise, they lead to the poor results. However in many practical situations it is often so difficult to determine the best number of clusters. On the other hand, there are other clustering algorithms such as Rival Penalized Competitive Learning (RPCL) or ISODATA which find the number of clusters automatically. But these clustering methods have a problem in locating the cluster centers appropriately. In this paper, a novel hybrid structure is proposed which uses these two types of clustering algorithms as complementary to improve clustering performance. Moreover, a new weighted version of RPCL algorithm which is called here WRPCL is suggested. This structure not only outperforms the choice of the optimal number of clusters but also improves the performance of the overall clustering result. © 2008 Springer-Verlag.