(Special session) Big-data in resource exploration and drilling data processing

[2201-08-05] Development of a machine learning–based method for characterizing the ore-forming processes in magmatic-hydrothermal systems

○Denghui Zhu^{1[Doctoral course]}, Jiajie Wang¹, Noriyoshi Tsuchiya^1,2 (1. Tohoku University, 2. National Institute of Technology, Hachinohe College)

Chairperson：久保 大樹 （京都大学）

Keywords:quartz, magmatic-hydrothermal system, machine learning, ore-forming process, identification

Quartz trace elements record physical and chemical evolutions of quartz growth, serving as vital indicators in exploring the evolution of ore-forming fluid and the precipitation of mineral deposits. However, previous studies utilizing quartz trace elements to discriminate ore-forming process in magmatic-hydrothermal systems only analyze two or three elements, which is ineffective in interpreting evolution process of the systems. Considering the promising advantages of machine learning technique for processing large-sized and high-dimensional data, this research aims to propose a new machine learning-based method to discern the ore-forming process in magmatic-hydrothermal systems based on quartz trace element chemistry.

In this study, we compiled 4000 quartz analysis data of 7 trace elements (Li, Na, Al, Ti, Ge, Sr, Sn) of different genetic environments from I-type porphyry-epithermal, S- and A-type granite-greisen systems by previous literatures. Then, by employing Random Forest to analyze the compiled quartz trace element content data, combined with Recursive Feature Elimination for selecting the key elements, we built different models to identify the forming environments of quartz in above mentioned systems step by step with the average F1 score higher than 95%. The roles of quartz trace elements in discriminating the evolution processes of different magmatic-hydrothermal systems were then summarized. Finally, Uniformed Manifold and Uniform Manifold Approximation and Projection was applied in visualizing the identification results. This research demonstrates the superior performance of our machine learning models and we expect our work can provide new insights into exploring ore-forming processes in magmatic-hydrothermal systems and valuable guidance for mineral exploration efforts, with potential applications in other geological settings.