2nd Scientific Day of School of Science and Technology, UNICAM

Given the compositional unchangeable characteristics over human time obsidians are one of the most preferred indicator of exchanges and interactions between prehistoric people. The chemistry of archaeological obsidians allows to investigate their origin in terms of volcanic district of provenance among the suitable geological sources of the Mediterranean. A large amount of chemical data reported in the literature concerning obsidians from the main different volcanic districts have been used in this project (Mt. Arci. Lipari. Palmarola, Pantelleria, Aegean, Carpathian). A chemical database of about 20,000 abundances over 45 variables, among major and trace elements, has been built, collected by about 35 different papers. Then, proper datasets have been created, choosing a significant number of samples from the different geological origins, described by several variables (predictors). selected among the most discriminant element abundances, with the aim to develop predictive multivariate provenance models. The most interesting elements for the provenance studies, suitable to represent chemical fingerprint of every source, are usually in trace amounts in obsidians and this is a limitation. which requires the use of different trace analytical methods (INAA, XRF, LA-ICP-MS, PIXE). producing a major source of uncertainty when comparing different data. This limitation is less severe when we consider major elements. whose data resulted generally more abundant and coherent, but less characteristic of the different sources respect to the minor ones. In general only trace or major elements whose variances among the geological sources resulted greater than the variances within the single sources have been submitted to different approaching in multivariate analysis. Their choice depended; heuristically, on the different sources to be distinguished and classified. Principal Component Analysis (PCA). Linear Discriminant Analysis (LDA): Soft Independent Modeling of Class Analogy (SIMCA]: Partial Least Squares Discriminant Analysis (PLS-DA) were the exploited techniques giving us the most coherent predicted results.