Abstract Basalts are ubiquitous mafic rocks found within diverse tectonic settings on Earth. Despite concerted efforts to distinguish tectonic settings through the chemical compositions of basalt, some features of these rocks related to tectonic processes can be obscured by weathering and erosion over geological time, making the discrimination results ambiguous. In this study, we utilized major and trace element data of clinopyroxene in basalts from five distinct tectonic settings: continental within-plate basalts (WPB), island arc basalts (IAB), ocean island basalts (OIB), oceanic floor basalts (OFB), and continental flood basalts (CFB). Employing three machine learning techniques—Multi-layer Perceptron (MLP), Support Vector Machine (SVM), and Random Forest (RF)—we aim to discriminate tectonic settings and magma affinities. Sparse Multinomial Regression (SMR) approach is used to quantitatively discern geochemical signatures that are distinctive of each tectonic setting. The outcomes reveal the efficacy of SVM, which attains an accuracy of 92.1% (major element-based) and 95.2% (major and trace elements-based) for tectonic discrimination. Furthermore, SVM achieves an accuracy of 92.9% (major elements-based) and 95.7% (major and trace elements-based) for magma affinity discrimination. Our study shows that the integration of electron microprobe data from clinopyroxene with machine learning techniques provides an effective approach to distinguish various tectonic settings and magma affinities. The classifier models have been applied to investigate the Neoproterozoic geodynamics of the Jiangnan Orogen. The models show the two pulses of mantle plume events (∼830 Ma and ∼785 Ma) triggering break-up of the Rodinia supercontinent.
Read full abstract