The importance of this work is indicated by the large number of techniques that are generally used to determine a natural classification structure in the data. The authors analyze six hierarchical, agglomerative techniques (single linkage, complete linkage, group average, weighted average, centroid, median) generating random observations (having uniform and normal distributions); then they apply clustering methods to them. Usually the random input data are generated with a given classification structure. However, in this study, the random input data do not have any structure, so a selection method is bad if it finds a “good classification structure” for this data. The authors conclude that the complete linkage method is the best, in accordance with other results from the literature.

Representative references are cited. Since the “comparison criterion” does not depend on the “dissimilarity measures,” this reviewer cannot express a preference for one of these algorithms. This study should be extended by increasing the volume of random data and considering other kinds of random data sets.