Density Based Cluster Validity Measurement for Fuzzy Clustering
Lingkui Meng, Chunchun Hu, Fengqiu Wang · 2006
Cluster validity index is used to evaluate the clustering result yielded by the fuzzy clustering algorithm. In this paper, a new cluster validity index is proposed to determine the optimal fuzzy c-partition produced by the fuzzy c-means algorithm. The proposed index introduces two evaluation factors: distribution density and uncertainty. The first factor measures the extent of closeness or compactness of the members within a cluster, and the second estimates the reliability of the results of fuzzy c-partition. A good fuzzy c-partition is expected to have a large distribution density and a low uncertainty degree. The experimental results based on three various data sets indicate that the proposed index is effective and efficient comparing with some existing validity indices. Especially, for the spatial data set, the proposed index can yields the better result