Geochemical Signatures and Element Interactions of Volcanic-Hosted Agates: Insights from Interpretable Machine Learning

To unravel the link between agate geochemistry, host volcanic rocks, and ore-forming processes, this study integrated elemental correlation analysis, interaction interpretation, and interpretable machine learning (LightGBM-SHAP framework with SMOTE and 5-fold cross-validation) using 203 in-situ element datasets from 16 global deposits. The framework achieved 99.01% test accuracy and 97.4% independent prediction accuracy in discriminating host volcanic rock types. Key findings reveal divergence between statistical elemental correlations and geological interactions. Synergies reflect co-migration/co-precipitation, while antagonisms stem from source competition or precipitation inhibition, unraveling processes like stepwise crystallization. Rhyolite-hosted agates form via a “crust-derived magmatic hydrothermal fluid—medium-low salinity complexation—multi-stage precipitation” model, driven by high-silica fluids enriching Sb/Zn. Andesite-hosted agates follow a “contaminated fluid—hydrothermal alteration—precipitation window differentiation” model, controlled by crustal contamination. Basalt-hosted agates form through a “low-temperature hydrothermal fluid—basic alteration—progressive mineral decomposition” model, with meteoric water regulating Na-Zn relationships. Zn acts as a cross-lithology indicator, tracing crust-derived fluid processes in rhyolites, feldspar alteration intensity in andesites, and alteration timing in basalts. This work advances volcanic-agate genetic studies via “correlation—interaction—mineralization model” coupling, with future directions focusing on large-scale micro-area elemental analysis.