Classification of Pottery Fragments Described by Concentration of Chemical Elements

In order to distinguish between archaeological pottery produced in the Etruscan city of Tarquinia and pottery produced in other coeval sites, we tested several supervised learning algorithms for classification. Pottery sherds were analised by X-ray fluorescence analysis and described in the dataset by the relative concentration of nine chemical elements. The dataset was unbalanced with about one fourth of negative samples, and contained repeated measures for each fragment; the number of repeated measures for each fragment ranged between two and seven. We carried out two types of experiments which differ in the way the repeated measures are exploited. The best performing models showed good performance, in terms of accuracy, sensibility and specificity.