Abstract
A substantial fraction of global gold production originates from unregulated or illegal operators, which generates severe environmental damage, threatens human rights, and could destabilize local and even national economies. Demonstrating the true origin of gold, with a geochemical fingerprint, remains a major challenge, particularly when similar sources are mixed. To address this issue, we apply the Uniform Manifold Approximation and Projection (UMAP), a nonlinear dimensionality-reduction method designed for analyzing high-dimensional datasets. In this study, gold compositional data were expressed as probability vectors and compared using Hellinger distance, enabling the visualization and clustering of samples across different stages of the production chain. Our analysis demonstrates that while beneficiation processes alter the absolute concentrations of certain elements, distinctive geochemical signatures are retained between natural gold samples and manufactured products. This persistence allows UMAP to reveal meaningful patterns of similarity and distinction, even when traditional methods fail to differentiate mixed or transformed materials. The results demonstrate that UMAP is a robust, powerful tool to support traceability and enhance measures against illegal gold mining. By strengthening the scientific basis for source attribution, this approach provides an innovative workflow to support regulatory frameworks and judicial actions aimed at disrupting illicit gold trade and promoting responsible resource management.