Cluster Tendency Assessment for Fuzzy Clustering of Incomplete Data

Abstract

The quality of results for partitioning clustering algorithms depends on the assumption made on the number of clusters presented in the data set. Applying clustering methods on real data missing values turn out to be an additional challenging problem for clustering algorithms. Fuzzy clustering approaches adapted to incomplete data perform well for a given number of clusters. In this study, we analyse dierent cluster validity functions in terms of applicability on incomplete data on the one hand. On the other hand we analyse in experiments on several data sets to what extent the clustering results produced by fuzzy clustering methods for incomplete data reect the distribution structure of data.