Several machine learning approaches have been developed for the identification of geochemical populations. In these approaches, the geochemical elements are usually the sole quantitative variables used as inputs for geochemical population recognition. This means that the presence of other qualitative variables, such as geological information, is overlooked in the analysis. Hierarchical clustering, as an unsupervised machine learning method, is a common approach for dimensional reduction in the analysis of geochemical data. In this study, an alternative to this technique, known as geostatistical hierarchical clustering (GHC), is applied to identify geochemical populations in 3D in the Bondar Hanza copper porphyry deposit, Iran. In this paradigm, the qualitative geological variables can also be incorporated for geochemical population identification, in addition to qualitative geochemical elements. In this study, an innovative solution is presented to tune the weighting parameters of each variable in GHC, based on the associations that the clusters (i.e., geochemical populations) should have with the geological information. The results are compared with k-means and number–size fractal/multifractal (N–S) methods. As a result, GHC showed better agreement with alterations, rock types, and mineralization zones in this deposit. Finally, some important instructions are provided for further mineral exploration.

已经开发了几种机器学习方法来识别地球化学种群。在这些方法中，地球化学元素通常是唯一的定量变量，用作地球化学种群识别的输入。这意味着在分析中忽略了其他定性变量的存在，例如地质信息。层次聚类作为一种无监督的机器学习方法，是地球化学数据分析中常用的降维方法。在这项研究中，该技术的替代方法，称为地质统计层次聚类 (GHC)，用于识别伊朗 Bondar Hanza 铜斑岩矿床的 3D 地球化学种群。在这个范式中，定性地质变量也可以被纳入地球化学种群识别，除了定性地球化学元素。在这项研究中，基于集群（即地球化学种群）与地质信息应具有的关联，提出了一种创新的解决方案来调整 GHC 中每个变量的加权参数。将结果与 k 均值和数字大小分形/多重分形 (N-S) 方法进行比较。因此，GHC 与该矿床的蚀变、岩石类型和矿化带表现出更好的一致性。最后，为进一步的矿产勘探提供了一些重要的说明。将结果与 k 均值和数字大小分形/多重分形 (N-S) 方法进行比较。因此，GHC 与该矿床的蚀变、岩石类型和矿化带表现出更好的一致性。最后，为进一步的矿产勘探提供了一些重要的说明。将结果与 k 均值和数字大小分形/多重分形 (N-S) 方法进行比较。因此，GHC 与该矿床的蚀变、岩石类型和矿化带表现出更好的一致性。最后，为进一步的矿产勘探提供了一些重要的说明。