Geochemical Discrimination of Agate in Diverse Volcanic Host Rocks Through Machine Learning Methods

The formation of agate is related to syngenetic or epigenetic magmatic hydrothermal fluids. Trace elements incorporated during the alteration of host rocks caused by hydrothermal magmatic fluids precipitate during their crystallization, reflecting the characteristics of the host rocks. While prior research has yielded extensive trace element data to differentiate agate types found in volcanic rocks, there remains a need for a more efficient method to identify the host volcanic rock of alluvial agate. In this study, a two-stage Random Forest approach was employed to classify the chemical compositions of agate originating from rhyolite, andesite, and basalt, with the aim of facilitating the determination of the host volcanic rock for unknown alluvial agate samples. A dataset comprising 203 agate compositional analyses, sourced from 16 distinct locations, was compiled and labeled for the purpose of training the Random Forest model. The classification results indicate that the developed models exhibit high accuracy (0.9524) and an F1 score of 0.9512, demonstrating their superior performance and efficiency. The feature importance analysis of these models reveals that U, Sb, and Sr contents are particularly crucial for discriminating between different types of volcanic rocks hosting agate. Furthermore, this study introduces a novel discriminant plot utilizing linear discriminant analysis (LDA) to evaluate the host volcanic rock of agate based on trace element data. Verifying the trace element data of agate samples related to basalt based on actual measurements shows that both the Random Forest (with accuracy of 0.813) and LDA plot underscore the effectiveness of using the trace elements found in alluvial agate for the identification of the host volcanic rock.