The global demand for critical minerals is surging, yet the challenge of discovering new deposits with traditional methods casts uncertainty on future supplies. This study introduces a machine learning-based framework to address common exploration challenges, such as limited known occurrences, difficulty in selecting negative samples, and unbalanced training data. By integrating an enhanced generative adversarial network with positive and unlabelled learning, the framework aims to improve exploration efficiency. Applied to the Gawler Craton in South Australia for cobalt, chromium, and nickel exploration, the framework generates prospectivity maps with a high spatial correlation to known occurrences, identifying potential areas for future exploration. The approach outperforms traditional methods, highlighting the importance of geophysical features in locating critical minerals and suggesting a more accurate and efficient way to pinpoint prospective mining regions.
@article{Farahbakhsh2023,
title = {Prospectivity modelling of critical mineral deposits using a generative adversarial network with oversampling and positive-unlabelled bagging},
author = {Farahbakhsh, Ehsan and Maughan, Jack and M{\"u}ller, R. Dietmar},
year = {2023},
journal = {Ore Geology Reviews},
number = {105665},
doi = {10.1016/j.oregeorev.2023.105665},
}