Data and code for: Machine-learning tectonic discrimination with uncertainty control in a large igneous province–subduction overlap zone

This record contains the data and code underlying the manuscript "Machine-learning tectonic discrimination with uncertainty control in a large igneous province–subduction overlap zone: Permian–Triassic mafic rocks of the Youjiang Basin, SW China" (manuscript submitted). It includes: (1) the compiled four-class training datasets (ocean-island basalt, volcanic arc basalt, continental-rift basalt and the Emeishan Large Igneous Province reference set) and the step-by-step ELIP screening record, derived from the GEOROC precompiled collections; (2) the prediction inputs for Permian–Triassic mafic rocks of the Youjiang Basin, none of which were used for training; (3) the Python modelling pipeline (21 scripts run in numerical order) covering preprocessing, grouped cross-validation, model training and consensus, open-set rejection, SHAP interpretation and the Siberian Traps validation; (4) per-sample and per-intrusion model outputs; (5) the scripts that draw Figs. 3–13 and Supplementary Fig. S1; and (6) Supplementary Tables S1–S12. File-level provenance and the model-data construction records are documented in the README. Note that the data files and scripts keep the internal class key "IAB" for the arc class called VAB in the manuscript (see README). The raw GEOROC source data are not redistributed; they were obtained from the precompiled collections Ocean Island Groups (https://doi.org/10.25625/WFJZKY), Convergent Margins (https://doi.org/10.25625/PVFZCE), Rift Volcanics (https://doi.org/10.25625/KAIVCT) and Continental Flood Basalts (https://doi.org/10.25625/WSTPOX). Data are released under CC BY 4.0 and code under the MIT Licence.