Recently, machine learning methods have been utilized to mine correlations between geological variables and mineral deposits because of their significance in mineral prospectivity mapping (MPM). However, the characteristics of known mineral deposits are often overlooked in supervised approaches to MPM because only spatial coordinates of known mineral deposits are used as positive training samples. We propose an interpretable method using association rules to predict mineral prospectivity in the Pangxidong district by incorporating characteristics associated with mineral deposits into MPM. Specifically, association rules are a type of data-driven equivalent of ore-controlling factors in knowledge-based exploration and it warrants a broader consideration in modern data-centric exploration. The detailed procedures are as follows: (1) two strong association rules related to mineral deposits were extracted using the Apriori algorithm based on the known Ag–Au and Pb–Zn deposits in Pangxidong; (2) the weights of the variables in the data filtered by the strong association rules were defined using entropy weight method (EWM); and (3) the probability of finding undiscovered mineral deposits was calculated. The Apriori algorithm delineated 57.3% and 52.6% of the known Ag–Au deposits and Pb–Zn deposits within 3.91% and 1.48% of the study area, respectively. In addition, after the EWM, high-probability areas of Ag–Au deposits and Pb–Zn deposits cover 1.05% and 0.43% of the study area, respectively. Therefore, the proposed method is effective and efficient in MPM and it has the potential to be applied more broadly.

最近，机器学习方法已被用来挖掘地质变量和矿床之间的相关性，因为它们在矿产前景图（MPM）中具有重要意义。然而，在 MPM 的监督方法中，已知矿床的特征经常被忽视，因为只有已知矿床的空间坐标被用作正训练样本。我们提出了一种可解释的方法，通过将与矿床相关的特征纳入 MPM，使用关联规则来预测庞溪洞地区的矿产前景。具体来说，关联规则是一种数据驱动的等价物，相当于基于知识的勘探中的控矿因素，它值得在现代以数据为中心的勘探中得到更广泛的考虑。具体流程如下：（1）根据庞溪洞已知的银金矿和铅锌矿床，利用Apriori算法提取两条与矿床相关的强关联规则； (2)采用熵权法(EWM)定义强关联规则过滤后的数据中变量的权重; (3)计算发现未发现矿藏的概率。 Apriori 算法分别在研究区域的 3.91% 和 1.48% 范围内描绘了 57.3% 和 52.6% 的已知银金矿床和铅锌矿床。此外，EWM之后，Ag-Au矿床和Pb-Zn矿床的高概率区域分别覆盖了研究区域的1.05%和0.43%。因此，所提出的方法在 MPM 中是有效且高效的，并且具有更广泛应用的潜力。