Machine Learning for Trace Elements and REE pattern Recognition and Typification with Dimensional Compression for Metal Traceability.

Abstract

This study aims to develop traceability technology by visualizing and typifying the trace elements and Rare Earth Elements (REEs) concentration patterns in ore minerals and final products by dimensionality reduction using machine learning. While the REE patterns of the two tungsten minerals, Scheelite and Wolframite, appear similar with no clear differentiation, the application of the LASSO-UMAP method for dimensionality reduction successfully distinguishes the REE patterns between these two minerals. On the other hand, trace elements and REEs in the final product form a nearly constant cluster, exhibiting no significant differences when analyzed with machine learning. This suggests that despite the various REE patterns in ore minerals from diverse mining locations, they eventually converge to a specific position on the UMAP. The trajectory from ore minerals to final products on the UMAP holds potential for application in metal traceability.