Data mining is a promising new tool in mineral exploration. Here, we combined data-mining procedures with spatial prediction modeling for gold exploration targeting in the Buhweju area in southwestern Uganda. It was employed in a data-rich context of unavoidably partly redundant and correlated information that offered challenges in extracting significant relationships. Our study utilized a database of co-registered digital maps related to gold mineralization. It comprised Landsat TM, Shuttle Radar Topographic Mission (SRTM), and geophysical (radiometric and magnetic) datasets for geological and structural mapping. The locations of 15 orogenic gold deposits and 87 gold occurrences were obtained from the Geological Survey of Uganda database. These were considered direct evidence of the presence of gold mineralization. The geological and geophysical settings at the gold deposit/occurrences locations were based on geological units as host rocks, contacts, and structural elements, together with continuous field values of geophysics, radiometry, and other remotely sensed imagery. A gold exploration targeting proposition (T_p) was defined as: “That a point p within the study area contains a gold deposit given the presence of spatial evidence.” All outstanding combinations of spatial evidence were obtained using empirical likelihood ratios. With a data-mining strategy, the ratios were filtered and modeled to identify stronger spatial associations, to rank the study area according to the likelihood of future discoveries, to represent ranking quality, to estimate associated uncertainty, and to select prospective target areas. The empirical likelihood ratios facilitated a transparent strategy for generating prediction patterns and extracting small prospective target areas with higher likelihood of discovery and lower-ranking uncertainty. Conclusions are provided on the knowledge extraction for prospectivity with further data and the challenges of reducing the arbitrariness of decisional steps.

数据挖掘是矿产勘探中一种很有前途的新工具。在这里，我们将数据挖掘程序与空间预测建模相结合，以在乌干达西南部的 Buhweju 地区进行黄金勘探。它被用于数据丰富的环境中，不可避免地存在部分冗余和相关信息，这为提取重要关系带来了挑战。我们的研究利用了与金矿化相关的共同注册数字地图数据库。它包括 Landsat TM、航天飞机雷达地形任务 (SRTM) 以及用于地质和结构测绘的地球物理（辐射和磁）数据集。从乌干达地质调查局数据库中获得了 15 个造山金矿床和 87 个金矿的位置。这些被认为是金矿化存在的直接证据。金矿/矿点位置的地质和地球物理设置基于作为主岩、接触点和构造元素的地质单元，以及地球物理、辐射测量和其他遥感图像的连续场值。黄金勘探目标主张（T _p ) 被定义为：“鉴于空间证据的存在，研究区域内的点p包含金矿床。” 所有突出的空间证据组合都是使用经验似然比获得的。通过数据挖掘策略，对比率进行过滤和建模以识别更强的空间关联，根据未来发现的可能性对研究区域进行排名，代表排名质量，估计相关的不确定性，并选择预期的目标区域。经验似然比促进了生成预测模式的透明策略并提取具有较高发现可能性和较低等级不确定性的小型预期目标区域。提供了关于前景知识提取的结论以及进一步的数据和减少决策步骤任意性的挑战。