A comparative study of cluster validation indices applied to genotyping data

Clustering is the most important task in unsupervised learning and cluster validation plays a very important role in cluster analysis. In this paper, we compared the performance of 7 major validation indices designed for Fuzzy-c Means: Partition Coefficient (PC), Partition Entropy (PE), Fukuyama-Sugeno index (F-S), Xie and Beni index (X-B), Compose Within and Between scattering (CWB), SC and Fuzzy hyper volume (FHV) on genotyping data obtained from single nucleotide polymorphism analysis. We first find there are three factors (the fuzzy factor m, the number of variables p and the maximum number of clusters cmax) that may influence validation indices' performance. A validation scheme was designed to optimize the performance of these indices. Finally, we test the indices on a total of 18 datasets and compared their performance. PC and CWB showed the best overall performance. CWB only failed on one dataset and PC failed on 2.