High-accuracy mineralization evaluation of VMS deposits using machine learning and basalt geochemistry

• Published in: Ore geology reviews Vol. 184; p. 106780
• Main Authors: Li, Jiachen, Sun, Xiang, Xiao, Ke, Wang, Qiuyun, Liang, Xiaoya, Cui, Limeng
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2025
• Subjects: AdaBoost; Basalt; Machine learning; Mineralization evaluation; VMS deposit; Basalt; AdaBoost; Mineralization evaluation; VMS deposit; Machine learning
• ISSN: 0169-1368
• DOI: 10.1016/j.oregeorev.2025.106780

[Display omitted] •Developed AdaBoost, GBDT, and RF models for VMS deposit evaluation, achieving over 99.63% accuracy with a basalt dataset.•Used SHAP analysis to identify Fe2O3, TiO2, and Co as critical indicators, linking ML results to geological processes.•Constructed "Co/Fe2O3 vs. V/Tm" and "Fe2O3 vs. TiO2" diagrams for visual, quantitative mineralization assessment. Basalt is a common volcanic rock in volcanogenic massive sulfide (VMS) deposits, and its geochemical composition provides critical insights into magmatic source characteristics, thereby serving as an essential proxy for evaluating the mineralization potential of VMS deposits. However, traditional assessment approaches often suffer from low efficiency due to the lack of clearly defined geochemical indicators and an overreliance on empirical interpretations. To address these limitations, we compiled a comprehensive global database of geochemical data for both mineralized and unmineralized basalts, and applied three machine learning algorithms—Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and Random Forest (RF)—to develop predictive models for VMS mineralization potential. All three models yielded high prediction performance, with both accuracy and F1-scores exceeding 99.63 %. Among them, the AdaBoost model achieved the best results, with an accuracy and F1-score of 99.79 %. Despite the strong predictive capabilities of these models, their “black-box” nature often limits the interpretability of feature importance. To enhance model transparency, we employed SHapley Additive exPlanations (SHAP) to quantify the contribution of each geochemical variable and to construct geochemically meaningful discrimination diagrams. The effectiveness of these indicators was further validated through logistic regression analysis. Our results indicate that Fe2O3, TiO2, and Co are among the most influential elements for distinguishing barren from fertile basalts. We developed classification diagrams based on key element ratios, notably Co/Fe2O3 vs. V/Tm and Fe2O3 vs. TiO2, which yielded classification accuracies of 95.51 % and 84.90 %, respectively. These diagrams offer intuitive and effective tools for rapid assessment of VMS mineralization potential. Overall, this study establishes a novel framework for objective, data-driven mineralization evaluation in VMS exploration.