AUC area of different machine learning models.

Wu-Zheng-Dao District in China is the world’s most famous mining areas. It hosts several world-class deposits, such as Xinming, Datang and Luolong bauxite deposits. Although this area still has significant potential for the discovery of new deposits, mineral prediction has become increasingly diffcult as the number of shallow deposits diminishes. Therefore, it is necessary to explore new and effective metallogenic prediction methods.Weights of evidence and machine-learning algorithms were used for mineral prospecting in this study. This study used a confusion matrix, receiver operating characteristic (ROC) curve,and prediction efficiency curve to evaluate the prediction results of each machine algorithm. The results showed that 95.9% of the deposits were located in high and distant scenic areas, accounting for 10% of the total area.The prospectivity map of the Wu-Zheng-Dao district shows that the high prospective areas are generally confined to the claystone and carbonatite rocks of the Eastern region, in particular, of the clay layers, and several areas of high prospectivity also occur in the Southern Cross Domain. According to the predicted results, after on-site exploration, design, and construction, Yanfengqian bauxite deposit was discovered, with an average thickness of 1.82 meters; The average content of Al2O3 is 61.24%; The resource amount is 28.9503 million tons.

中国吴正道矿区是全球最著名的矿区之一，赋存新明、大唐、洛隆等世界级铝土矿矿床。尽管该区域仍具备发现新矿床的巨大潜力，但随着浅部矿床数量日渐减少，矿产预测工作的难度也日益攀升。因此，探索新型高效的成矿预测方法迫在眉睫。
本研究采用证据权法（Weights of Evidence）与机器学习算法开展矿产勘查工作。本研究通过混淆矩阵（Confusion Matrix）、受试者工作特征（Receiver Operating Characteristic, ROC）曲线以及预测效率曲线，对各机器学习算法的预测结果进行评估。研究结果表明，95.9%的矿床坐落于占研究区总面积10%的高潜力远景区内。
吴正道矿区的成矿预测图显示，高潜力成矿区主要集中于东部区域的泥岩与碳酸盐岩地层中，尤以黏土层最为显著；此外，南十字域也分布有多处高潜力成矿区。基于预测结果，经野外勘查、方案设计与工程施工后，研究团队成功探明岩峰前铝土矿矿床，其平均厚度为1.82米，三氧化二铝（Al₂O₃）平均品位为61.24%，资源量达2895.03万吨。