Determining the origins of Phoenician silver: Exploring the potential of machine learning for lead isotope analysis

This study investigates the provenance of Phoenician silver hoards through the integration of machine learning algorithms with lead isotope analysis (LIA). Utilizing an extensive isotopic database of ores, advanced clustering techniques (DBSCAN) enabled the identification of distinct groups within the data. Outliers were removed, and dataset imbalances were addressed using the SMOTE algorithm, thereby enhancing the prediction accuracy. An efficient and robust machine learning model (XGBoost) was employed to classify isotopic signatures and assign probable ore sources with high confidence. To test the model, endmembers from the previously published Southern Levantine Iron Age silver hoards of Tel Dor, ʿAkko, ʿEin Hofez, and ʿArad were defined by using a new automated algorithm for endmember selection. The algorithm applies principal component analysis (PCA) aligned with the geological age model (geochron) to define isotopic endmembers in a structured and reproducible way, moving beyond purely visual identification. Although exploratory in nature, and with some manual adjustments still required, this approach offers a more systematic framework for provenance studies. Our findings support previous interpretations regarding early Phoenician endeavors to Sardinia and Anatolia, highlight the role of Iberian silver, and introduce new possible connections to the Aegean. Overall, this research demonstrates the potential of combining automated endmember identification with machine learning to provide new insights into ancient Mediterranean trade networks. • Machine learning applied to Phoenician silver lead isotope analysis. • DBSCAN clustering used to detect outliers and subgroup patterns. • SMOTE balancing and XGBoost improved provenance classification. • Endmember identification reveals composite metal source scenarios. • Transparent scalable framework for future provenance studies in archaeometallurgy.