A new validity index based on intra-cluster variation and inter-cluster overlap
Guangda Su · Journal of Optoelectronics·laser · 2010
The determination of cluster number is still an open problem for fuzzy C-means clustering.In this paper,a new validity index is proposed to evaluate partition and determine the optimal number of clusters for fuzzy clustering.In a good partition,the similarities of patterns in a cluster should be maximized and the clusters should be well separated.Intra-cluster variation and intercluster overlap are defined to measure the similarities within a cluster and the separation between clusters respectively.The validity index is defined based on the two measurements.Experimental results on four artificial datasets and two real datasets show the effectiveness and robustness of the proposed validity index.