Scholarlink

<p>As surface and near-surface gold resources become increasingly depleted, global gold exploration is progressively targeting deeper, concealed ore bodies. This shift has substantially elevated exploration challenges, necessitating the application of novel approaches to enhance success rates while mitigating risks and costs. This study compiled a comprehensive global geochemical dataset (2642 effective samples) of magmatic rocks, characterized by major elements (SiO<sub>2</sub>, TiO<sub>2</sub>, Al<sub>2</sub>O<sub>3</sub>, FeOt, MnO, MgO, CaO, Na<sub>2</sub>O, K<sub>2</sub>O, P<sub>2</sub>O<sub>5</sub>) and trace elements/ratios (Rb, Ba, Nb, Sr, Zr, Ba/Zr, Nb/Zr) genetically linked to porphyry, skarn, epithermal, and reduced intrusion-related gold mineralization. After comparing four dimensionality reduction techniques (PCA, t-SNE, Isomap, and UMAP), UMAP was selected for its ability to preserve both local and global structures in high-dimensional data and produce more compact clustering. Subsequently, two machine learning algorithms—weighted random forest (WRF) with class weighting and support vector machine (SVM) with RBF kernel—were applied to a training dataset. Two hybrid models were constructed for predicting the gold fertility of magmatic rocks: UMAP + WRF and UMAP + SVM. The UMAP + WRF model achieved higher training accuracy (0.9735) compared to UMAP + SVM (0.9529). Validation on a geographically independent dataset from the Central Andes demonstrated superior performance of the UMAP + WRF model across all metrics (precision, accuracy, recall, AUC(ROC), and AUC(PR)), confirming its greater suitability for practical mineral exploration prediction. A three-level visualization analysis elucidated the relationship between UMAP dimensions and geochemical indicators, showing patterns consistent with established geological models, in which thick crust and the involvement of subducted materials and fluids in the source region are important factors associated with magmatic-hydrothermal gold mineralization, thereby supporting the geological validity of the model. Finally, an online web application was developed to facilitate the model's application, allowing users to input whole-rock geochemical data and obtain predictions of gold mineralization potential, thereby promoting the adoption of machine learning methods in magmatic-hydrothermal gold deposit exploration.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting Magmatic-Hydrothermal Gold Fertility: A Hybrid Machine Learning and Dimensionality Reduction Approach with Global Igneous Geochemistry

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Fengge Han,
  • Cheng-Biao Leng,
  • Jiajie Chen,
  • Mangen Li

摘要

As surface and near-surface gold resources become increasingly depleted, global gold exploration is progressively targeting deeper, concealed ore bodies. This shift has substantially elevated exploration challenges, necessitating the application of novel approaches to enhance success rates while mitigating risks and costs. This study compiled a comprehensive global geochemical dataset (2642 effective samples) of magmatic rocks, characterized by major elements (SiO2, TiO2, Al2O3, FeOt, MnO, MgO, CaO, Na2O, K2O, P2O5) and trace elements/ratios (Rb, Ba, Nb, Sr, Zr, Ba/Zr, Nb/Zr) genetically linked to porphyry, skarn, epithermal, and reduced intrusion-related gold mineralization. After comparing four dimensionality reduction techniques (PCA, t-SNE, Isomap, and UMAP), UMAP was selected for its ability to preserve both local and global structures in high-dimensional data and produce more compact clustering. Subsequently, two machine learning algorithms—weighted random forest (WRF) with class weighting and support vector machine (SVM) with RBF kernel—were applied to a training dataset. Two hybrid models were constructed for predicting the gold fertility of magmatic rocks: UMAP + WRF and UMAP + SVM. The UMAP + WRF model achieved higher training accuracy (0.9735) compared to UMAP + SVM (0.9529). Validation on a geographically independent dataset from the Central Andes demonstrated superior performance of the UMAP + WRF model across all metrics (precision, accuracy, recall, AUC(ROC), and AUC(PR)), confirming its greater suitability for practical mineral exploration prediction. A three-level visualization analysis elucidated the relationship between UMAP dimensions and geochemical indicators, showing patterns consistent with established geological models, in which thick crust and the involvement of subducted materials and fluids in the source region are important factors associated with magmatic-hydrothermal gold mineralization, thereby supporting the geological validity of the model. Finally, an online web application was developed to facilitate the model's application, allowing users to input whole-rock geochemical data and obtain predictions of gold mineralization potential, thereby promoting the adoption of machine learning methods in magmatic-hydrothermal gold deposit exploration.