The continuous growth in global demand for mineral resources and the increasing difficulty of mineral exploration have created bottlenecks for traditional mineral prediction methods in handling complex geological information and large amounts of data. This review aims to explore the latest research progress in machine learning technology in the field of large-scale mineral prediction from 2016 to 2025. By systematically searching the Web of Science core database, we have screened and analyzed 255 high-quality scientific studies. These studies cover key areas such as mineral information extraction, target area selection, mineral regularity modeling, and resource potential evaluation. The applied machine learning technologies include Random Forests, Support Vector Machines, Convolutional Neural Networks, Recurrent Neural Networks, etc., and have been widely used in the exploration and prediction of various mineral deposits such as porphyry copper, sandstone uranium, and tin. The findings indicate a substantial shift within the discipline towards the utilization of deep learning methodologies and the integration of multi-source geological data. There is a notable rise in the deployment of cutting-edge techniques, including automatic feature extraction, transfer learning, and few-shot learning. This review endeavors to synthesize the prevailing state and prospective developmental trajectory of machine learning within the domain of large-scale mineral prediction. It seeks to delineate the field’s progression, spotlight pivotal research dilemmas, and pinpoint innovative breakthroughs.

Gejiu, a prominent tin–polymetallic ore district, is distinguished by its diverse mineral complexes. However, the genesis of these complexes and their relationship with mineralization remain inadequately studied. This study utilized whole-rock geochemical analyses to investigate the magmatic sources and petrogenesis of different complex types, aiming to elucidate their implications for tin–polymetallic mineralization. The results indicate that gabbro, monzonite, diorite, and syenite are derived from enriched mantle-derived magmas and have undergone limited crustal contamination. Granites are formed by the mixing of mantle- and crust-derived magmas, involving both physical mixing and chemical diffusion. Major and trace element characteristics suggest that the Gejiu granites predominantly exhibit features of both A-type and I-type granites. Harker diagrams and whole-rock indicators, such as Nb/Ta and Zr/Hf, suggest that granites experienced a two-stage fractional crystallization process, ultimately forming highly evolved biotite monzogranite. Fractional crystallization is the dominant mechanism controlling magmatic evolution, while high-temperature melting and biotite decomposition reactions are critical for the formation of the world-class Gejiu tin deposit.

With the rapid development of the global new-energy industry, lithium, as a key strategic mineral resource, has faced increasingly prominent supply–demand contradictions. The Anning area in Central Yunnan serves as an important clay-type lithium resource base in China, and its mineralization prediction research holds significant theoretical value and practical implications. This study focuses on the lithium-rich claystone of the Daoshitou Formation in the Anning area of Central Yunnan. Based on an in-depth analysis of typical clay-type lithium mineralization characteristics, this research analyzes regional lithium metallogenic conditions, identifies key ore-controlling factors and prospecting indicators, comprehensively collects geological, geophysical, geochemical, and remote sensing data, and completes digital processing. By integrating multi-source geoscience data, a multi-source geoscience information spatial database was constructed. Using the Random Forest algorithm as the core method and comparing with machine learning models such as XGBoost and Support Vector Machine (SVM), a mineralization prediction model was established to achieve quantitative prediction of clay-type lithium mineralization probability. The specific research findings are as follows: (1) The metallogenic model and prospecting indicators of clay-type lithium deposits were clarified. The Anning area of Central Yunnan mainly develops sedimentary clay-type lithium deposits, and mineralization is jointly controlled by lithium-rich source material supply, chemical weathering under warm and humid paleoclimate conditions, and weak reductive sedimentary environments of coastal-swamp facies; (2) a multi-source geoscience information database was constructed, extracting 16 predictive variables and establishing a metallogenic prediction sample dataset; (3) the Random Forest model performed optimally with an AUC value of 0.84, accuracy of 88.3%, precision of 89.4%, recall of 87.2%, and F1 score of 0.88; (4) three metallogenic prospect areas were delineated. The results indicate that under the study area conditions and existing data, the Random Forest model demonstrates good applicability and stability. The research results provide important target areas for lithium mineral exploration and deployment in the Anning area of Central Yunnan, and offer methodological references for similar deposit prediction research.

Mineral prospectivity mapping of weathering-crust-type ilmenite placer deposits is challenged by the limited number of known mineral occurrences, the lack of reliable barren-site labels, and spatial autocorrelation. This study focuses on the Wuding–Luquan area of Yunnan Province, China, and develops an eight-predictor framework comprising elevation, slope, fault kernel density, an interpolated Ti-concentration predictor, FeO_x, Al–OH and Mg/Fe–OH mineral anomalies, and distance to mafic rocks. Forty-three known ilmenite occurrences were used as positive samples, and 86 pseudo-absence/background samples were generated under spatial constraints. Random forest, XGBoost, and a multilayer perceptron, together with an equal-weight soft-voting ensemble, were used for mineral prospectivity mapping. Model performance and predictor contributions were evaluated using nested spatial cross-validation, SHapley Additive exPlanations (SHAP), permutation importance, and spatial-block bootstrap resampling. The equal-weight ensemble achieved the highest out-of-fold ROC-AUC (0.9202) and average precision (0.8580), while yielding the lowest log loss and Brier score. Ti was the most important predictor across all four modeling schemes, whereas Al–OH mineral anomalies, slope, and Mg/Fe–OH mineral anomalies also showed relatively high contributions in the equal-weight ensemble. A total of 379 measured TiO₂ records showed weak positive and scale-dependent rank associations with predicted prospectivity, with median Spearman coefficients ranging from 0.133 to 0.325 across aggregation scales of 250–2000 m. This comparison was treated as an exploratory geochemical comparison rather than as independent validation of model performance. By integrating model probabilities with the distributions of mafic rocks, known mineral occurrences, faults, and measured TiO₂, eight prospective areas (T1–T8) were delineated. The results provide a quantitative basis for regional mineral prospectivity assessment and exploration-target prioritization in the Wuding–Luquan area.

Mineral resource exploration is crucial for national economic security and sustainable development. Conventional mineral prediction methods struggle with low efficiency, high costs, and insufficient accuracy in predicting concealed deep - seated deposits and those in complex tectonic areas. Recently, machine learning algorithms, with powerful data - mining and pattern - recognition capabilities, have greatly enhanced mineral prediction intelligence. This paper comprehensively sums up the research progress and development trend of machine learning in large - scale mineral resource prediction, covering algorithm frameworks like random forests, support vector machines, and deep learning, as well as data - preprocessing techniques such as feature engineering and multi - source data fusion. For instance, the random forest algorithm achieved an 82% target - area match rate in Tibet's Julong porphyry Cu - Mo deposit prediction, cutting exploration costs by about 40%; in Fujian's southwestern Makeng iron deposit study, comparing four algorithms including random forests and support vector machines, the optimal model got an AUC of 0.88. However, issues like uneven data quality and poor model interpretability remain to be solved. In the future, combining generative adversarial networks (GANs) to boost data diversity and developing a "geological constraint + AI - driven" interdisciplinary paradigm are needed to promote the intelligent and precise transformation of mineral resource exploration.