Classification of archaeological sherds across the southeast United States based on variable selection from compositional fingerprints

The transfer of advances in chemometrics into archaeometric research opens a wide range of new application possibilities in this rapidly developing field. The present research represents a feasibility study aimed at showing how the huge potential that multivariate analysis and feature selection techniques have demonstrated for classification purposes can be extrapolated to archaeological provenance studies, thus pursuing an enhancement of the resulting classification performance. The classification problem studied here was related to the discrimination of pottery sherds from different sources across the southeast of the United States from their compositional fingerprints. The sample elemental concentrations were analyzed using the stepwise linear discriminant analysis (SLDA) method, thus simultaneously performing feature selection and classification. Several approaches, more or less restrictive according to the geographical scope and the number of considered classes, were explored, following a hierarchical classification approach. In contrast to previous studies on the same data set, the reliable and unequivocal classification strategy presented here did not merely focus on developing a large-scale classification into broad geographical areas, but finer classifications were also successively obtained until samples were assigned into individual regions. The great discrimination ability and effectiveness of the classification methodology proposed are promising and encourage its application to new samples of unknown provenance and the feasibility of using similar approaches in other archaeological studies. The high quality results obtained were even more remarkable considering the relatively small number of discriminant variables selected in each case by the stepwise procedure.