Concerning traditional models and machine learning approaches in geochemical data processing

Recent advances in machine learning (ML) highlight significant improvements in judgment capabilities, particularly in geochemical sciences and environmental research. This study underscores the evolution from traditional geoscientific models to modern data-driven approaches. It introduces an application of ML methods including unsupervised learning and dimensionality reduction to identify geochemical characteristics in mineral compositions. Integrating mineral chemistry analysis with ML addresses inefficiencies in processing large datasets generated by electron probe micro-analyzers (EPMA). We introduce a new algorithm for automating mineral chemistry data processing, achieving approximately 90% accuracy in classifying minerals from an extensive dataset of ultramafic rocks. Peridotites are believed to form from the partial melting of the Earth's mantle and represent the residues left after the extraction of basaltic melts during mantle upwelling. Abyssal (oceanic lithosphere) and fore-arc (subduction zone) peridotites provide valuable insights into the composition and dynamics of the upper mantle, including melt generation and mantle convection processes, interactions between tectonic plates, and the formation of new oceanic crust. Discriminating between abyssal and fore-arc peridotite tectonic settings using geochemical diagrams has been challenge due to significant data overlap. Initial analyses with established frameworks based on various geochemical parameters yielded inconclusive results. However, a new method focusing on mono-mineral compositions of spinel, olivine, and pyroxenes—validated with data from 600 samples across multiple ophiolitic/oceanic locations—can discriminate the abyssal and fore-arc settings more successfully than the traditional methods. Therefore, ML techniques have been employed to enhance discrimination between the two settings for new investigations and/or for previously well-studied peridotites.