A study of Roman pottery (terra sigillata) using hierarchical fuzzy clustering

In this paper the analysis of a set of 48 Roman pottery sherds is described using divisive hierarchical fuzzy clustering algorithms. The fuzzy clustering algorithms are considered to be capable of eliminating the disfunctionalities of the clustering algorithms used in the article of Aruga et al. [Anal. Chim. Acta, 276 (1993) 197]. In the present article principal component analysis is described both to reduce the characteristics space from R7 to R3 and R2, and to order the seven characteristics. Two reduced sets of data are characterized: the set obtained by the space reduction and the set obtained by selecting the three more important characteristics. The results are consistent with those of the original set of data. The order of characteristics differs from that given in the literature. We may thus infer that the characteristics selected in the literature are not the best ones.