Performance Assessment of Some Clustering Algorithms Based on a Fuzzy Granulation-Degranulation Criterion

In this paper a fuzzy quantization dequantization criterion is used to propose an evaluation technique to determine the appropriate clustering algorithm suitable for a particlar data set. In general, the goodness of a partitioning is measured by computing the variances within it, which is a measure of compactness of the obtained partitioning. Here a new kind of error function, which reflects how well the formed cluster centers represent the whole data set, is used as the goodness of the obtained partitioning. Thus a clustering algorithm, providing a good set of centers which approximate the whole data set perfectly, is best suitable for partitioning that particular data set. Five well-known clustering algorithms, GAK-means (genetic algorithm based K-means algorithm), a newly developed genetic point symmetry based clustering technique (GAPS-clustering), average linkage clustering algorithm, expectation maximization (EM) clustering algorithm and self organizing map (SOM) are used as the underlying partitioning techniques. Five artificially generated and three real-life data sets are used to establish that the proposed methodology is able to correctly identify appropriate clustering algorithm for a particular data set.