A machine learning framework to lanthanide element distribution and predictability from the northwestern Iranian karst bauxite deposits

This study applies Extreme Gradient Boosting (XGBoost) to examine how geological age clustering affects the predict- ability of light rare earth elements (LREEs: La–Sm) and heavy rare earth elements (HREEs: Eu–Lu) in the northwestern Iranian karst bauxites. Major oxides and weathering indices (CIA and CIW) were used as predictors, revealing contrast- ing behaviors between the Paleozoic and Mesozoic deposits. LREEs exhibited high accuracy in the Mesozoic deposits, whereas HREEs achieved the best predictability in the Paleozoic samples, reflecting distinct mineralogical controls. Vari- ability in cerium, linked to paleoredox conditions, reduces LREEs model performance in the Paleozoic, while phosphate phases strongly influence HREEs enrichment. Feature importance consistently identifies P2O5 as the dominant predictor for both LREEs and HREEs across all deposits, highlighting the key role of phosphate minerals in lanthanide incorpo- ration, whereas major oxides contribute less. CIA and CIW further enhance predictive accuracy, indicating that subtle variations in paleoweathering conditions affect lanthanide distribution and model performance. The LREEs and HREEs prediction models demonstrate promising potential, and further cross-validation across global karst bauxite deposits could improve understanding of the factors controlling REEs distribution, ultimately supporting more efficient and cost-effective exploration strategies for these critical metals.