||Cluster analysis is one kind of data mining. Its main feature is “A cluster is a collection of data objects that are similar to one another within the same cluster and dissimilar to the objects in other clusters” [J. Han and M. Kamber, 2001]. Clustering is a useful method to discover the hidden information, so it is applied in many fields, such as DNA analysis, Business Intelligence (BI), Computing Intelligence, etc.
Fuzzy c-means (FCM) algorithm has been good performance, but it still has some drawbacks. For example, it must get the predefined number of clusters in advance as traditional k-means. However in the real world application, the cluster number usually can’t be known beforehand. Hierarchical clustering doesn’t need to know the number of clusters in advance, but its time complexity is very high. Researchers [M. Steinbach, G. Karypis, V. Kumar, 2000] offer the Bisecting k-means clustering to resolve the problem on cluster number and receive lower time complexity than original hierarchical clustering. But this method still suffers from the fact that once a step is done, it can never be undone which causes a big problem that some data been partitioned into wrong clusters in the previous stage won’t be relocated in following stages.
Therefore, the purpose of this research is to automatically select the optimal number of clusters and achieve high accuracy. This research integrates the property of bisecting and divisive hierarchical clustering with Fuzzy c-means algorithm, and finally merges the high similar clusters, which maybe wrong partitioned into different clusters, to improve the clustering quality. We present three enhanced relocatable split-oriented fuzzy clustering algorithms called Bisect FCM, Merging Bisect FCM and Split FCM. The experimental results show that our approaches are not only efficient in convergence speed but also effective in clustering quality. They can obtain the correct or near the cluster number and achieve even higher accuracy than the FCM with cluster number known beforehand.