Pyrite geochemistry is crucial for the discrimination of the types of ore-forming fluids in gold deposits, such as metamorphic–hydrothermal fluids and magmatic–hydrothermal fluids. With the assistance of supervised machine learning algorithms, this application can be leveraged maximally. Here, laser ablation inductively coupled plasma mass spectrometry (LA–ICP–MS) geochemical data for 4683 pyrite samples worldwide were collected to train seven classification models. The top three algorithms, including Random Forest (RF), Support Vector Machines (SVM), and Multilayer Perceptron (MLP), were used to build classifiers to predict the type of pyrite. The established classifiers were applied to new geochemical data for pyrite samples collected from the Jinkeng and Huanggou gold deposits in the Xuefengshan Orogen (XFSO). The findings suggest that the classifiers are capable of accurately distinguishing between two main types of ore-forming fluids, with good predictive outcomes. This performance surpasses that of traditional, two-dimensional diagram-based methods. The classifiers determined that the geochemical constituents of pyrite from the Jinkeng and Huanggou originated from metamorphic–hydrothermal sources, consistent with geological and geochemical evidence. The results further reveal that the Jinkeng and Huanggou are classified mostly as orogenic gold deposit. This study proves that data-driven methods based on machine learning can provide compelling evidence for distinguishing between the types of ore-forming fluids, understanding deposit genesis, and providing prospecting ideas. Additionally, this research boosts confidence in the use of machine learning to geological classification challenges.

黄铁矿地球化学对于金矿床中变质热液、岩浆热液等成矿流体类型的判别至关重要。在监督机器学习算法的帮助下，可以最大限度地利用该应用程序。在这里，收集了全球 4683 个黄铁矿样品的激光烧蚀电感耦合等离子体质谱 (LA-ICP-MS) 地球化学数据，以训练七个分类模型。排名前三的算法，包括随机森林（RF）、支持向量机（SVM）和多层感知器（MLP），被用来构建分类器来预测黄铁矿类型。所建立的分类器应用于从雪峰山造山带（XFSO）金坑和黄沟金矿采集的黄铁矿样品的新地球化学数据。研究结果表明，分类器能够准确区分两种主要类型的成矿流体，并具有良好的预测结果。这种性能超越了传统的基于二维图的方法。分类器确定金坑和黄沟黄铁矿的地球化学成分来源于变质热液源，与地质和地球化学证据一致。结果进一步表明，金坑和黄沟大多属于造山型金矿床。这项研究证明，基于机器学习的数据驱动方法可以为区分成矿流体类型、了解矿床成因、提供找矿思路提供令人信服的证据。此外，这项研究增强了人们对使用机器学习应对地质分类挑战的信心。