Machine learning-based approach for adakitic rocks tectonic setting determination

Adakitic rocks hold significant geodynamic and metallogenic implications, and accurately determining their tectonic setting is crucial for understanding regional tectonic-magmatic evolution. However, due to the diverse sources, heat regimes, and magma generation mechanisms of adakitic rocks, conventional low-dimensional geochemical methods face limitations in tectonic setting identification. With the exponential growth of geoscience data and advancements in artificial intelligence, machine learning offers a novel approach to address this challenge. In this study, we integrate machine learning with geological big data to develop a high-precision adakitic tectonic setting discrimination model and visual representation. We compiled major and trace elements geochemical data from 1075 adakitic rocks worldwide and employed unsupervised learning techniques such as principal component analysis and t-distributed stochastic neighbor embedding for high-dimensional data reduction. Various machine learning algorithms including random forest, support vector machine, artificial neural network, and K-nearest neighbor were trained. Consequently, we established a Gaussian kernel support vector machine adakitic rock tectonic setting discriminator with 98. 5% accuracy and proposed a Ba versus Sr/Nd diagram to delineate three tectonic settings: convergent margin, intraplate volcanism, and Archean craton (comprising greenstone belts). This study broadens the application of machine learning in adakitic rock tectonic setting analysis, offering fresh insights into tectonic-magmatic processes investigation. © 2024 Science Frontiers editorial department. All rights reserved.