Scholarlink

<p>The essence of mineral prospectivity prediction lies in the quantitative extraction and integration of multi-source geoscience information, with current research frontiers focusing on the intelligent extraction of ore-controlling factors and their nonlinear modeling. With the rapid development of data science and artificial intelligence, the integration of mathematical models and machine learning algorithms for multi-source mineral exploration data mining has emerged as a key research frontier in mineral prospectivity modeling (MPM). In this study, the box-counting method and the spectrum–area (S–A) fractal model were used in combination to quantitatively analyze structural features, remote-sensing-derived alteration information, and geochemical anomalies. The study revealed that high fractal dimension zones of remote-sensing-derived alterations are spatially consistent with alteration anomalies, while structural patterns identified through fractal analysis showed strong spatial agreement with known ore-controlling structures, offering robust alteration and structure related predictor variables for MPM. For geochemical information extraction, compositional data analysis was applied to construct principal component (PC1 and PC2) score maps, which were further combined with the S–A fractal model to identify corresponding geochemical anomaly fields. This integrated approach systematically revealed the spatial distribution patterns of principal components and the underlying geochemical anomaly structures, thereby providing essential geochemical predictors for model development. By integrating multi-source prospecting predictors, including metallogenic geological background, structural and remote sensing alteration fractal modeling results, and geochemical characteristics, three predictive models were developed: a convolutional block attention module-based convolutional neural network (CBAM-CNN), a simulated annealing-based random forest (SA-RF), and an improved particle swarm optimization-based support vector machine (IPSO-SVM). Following a comparative evaluation of model performance, the Shapley additive explanations (SHAP) method rooted in cooperative game theory was utilized to perform global interpretability analysis of the best-performing model, enabling a quantitative assessment of both the magnitude and direction of each mineral prospectivity predictor’s contribution to the model output. The results indicated that major ore-controlling structures, alteration-related fractal dimensions, and geochemical principal components were assigned higher explanatory weights in the model, effectively uncovering the dominant controlling factors and their coupling relationships in regional mineralization. These findings underscore the theoretical significance and practical utility of the SHAP method in data-driven geoscientific modeling.</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

Integrating Fractal Theory with Explainable Machine Learning for Mineral Prospectivity Mapping in the Nenjiang–Heihe Region

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Zhonghai Zhao,
  • Xiang Zhao,
  • Bo Zheng,
  • Chenglu Li,
  • Yang Liu,
  • Kaiyue Tan,
  • Yunbao Yang,
  • Zhaolong Liu,
  • Yechang Yin

摘要

The essence of mineral prospectivity prediction lies in the quantitative extraction and integration of multi-source geoscience information, with current research frontiers focusing on the intelligent extraction of ore-controlling factors and their nonlinear modeling. With the rapid development of data science and artificial intelligence, the integration of mathematical models and machine learning algorithms for multi-source mineral exploration data mining has emerged as a key research frontier in mineral prospectivity modeling (MPM). In this study, the box-counting method and the spectrum–area (S–A) fractal model were used in combination to quantitatively analyze structural features, remote-sensing-derived alteration information, and geochemical anomalies. The study revealed that high fractal dimension zones of remote-sensing-derived alterations are spatially consistent with alteration anomalies, while structural patterns identified through fractal analysis showed strong spatial agreement with known ore-controlling structures, offering robust alteration and structure related predictor variables for MPM. For geochemical information extraction, compositional data analysis was applied to construct principal component (PC1 and PC2) score maps, which were further combined with the S–A fractal model to identify corresponding geochemical anomaly fields. This integrated approach systematically revealed the spatial distribution patterns of principal components and the underlying geochemical anomaly structures, thereby providing essential geochemical predictors for model development. By integrating multi-source prospecting predictors, including metallogenic geological background, structural and remote sensing alteration fractal modeling results, and geochemical characteristics, three predictive models were developed: a convolutional block attention module-based convolutional neural network (CBAM-CNN), a simulated annealing-based random forest (SA-RF), and an improved particle swarm optimization-based support vector machine (IPSO-SVM). Following a comparative evaluation of model performance, the Shapley additive explanations (SHAP) method rooted in cooperative game theory was utilized to perform global interpretability analysis of the best-performing model, enabling a quantitative assessment of both the magnitude and direction of each mineral prospectivity predictor’s contribution to the model output. The results indicated that major ore-controlling structures, alteration-related fractal dimensions, and geochemical principal components were assigned higher explanatory weights in the model, effectively uncovering the dominant controlling factors and their coupling relationships in regional mineralization. These findings underscore the theoretical significance and practical utility of the SHAP method in data-driven geoscientific modeling.