Tracing Andean Origins: A Machine Learning Framework for Lead Isotopes

Abstract

This study uses machine learning techniques to facilitate the geolocation of Andean lead isotopes, a novel approach in this geographical context. Two predictive models for latitude and longitude were developed based on the compilation of a database of the lead isotope ratios 206Pb/204Pb, 207Pb/204Pb, and 208Pb/204Pb from multiple Andean provinces. These models were cross-validated using GridsearchCV to assess their performance, identifying Random Forest as the best-performing model. Also, clustering analysis with the K-means model and Euclidean distance was used to correlate artifact isotope compositions with known sources. The limitations and scope of the models were listed for their appropriate usability and interpretability. This work extends basic geochronological studies, integrates a comprehensive database, and applies state-of-the-art algorithms to generate predictive models, contributing to a deeper understanding of the historical Andean mineral resources' distribution and use.