Key points are not available for this paper at this time.
Semi-supervised learning warrants more significant consideration for machine learning-based mapping in mineral exploration , since mineral deposits frequently exhibit imbalances in occurrence frequencies. It can potentially address challenges associated with class imbalances via the efficient use of labeled data and the extrapolation of patterns from unlabeled data. This research endeavors to present a prospective model for Mississippi Valley-Type lead and zinc deposits employing a semi-supervised approach in the Varcheh district, western Iran. To achieve this goal, diverse exploratory criteria related to mineralization, encompassing geological, remote sensing , geochemical, and structural layers, have been incorporated to develop a semi-supervised mineral prospectivity model. The model strengthens the advantages of supervised and unsupervised learning approaches by incorporating the Categorical gradient Boosting (CatBoost) and Gaussian mixture model algorithms into a semi-supervised framework. This approach effectively utilizes limited labeled data, while capturing spatial patterns and relationships in the unlabeled dataset, ultimately contributing to a more robust mineral prospectivity mapping model. Indeed, the regions with high posterior probability include most lead and zinc deposits in this strategy, suggesting that the locations of known deposits are significantly tied to areas connected to high posterior probability. The semi-supervised proposed framework in this paper is also compared with supervised approach to validate the performance improvement. The implemented approach can be highly valuable for exploring resources. • Revolutionizing mineral exploration with semi-supervised learning in imbalanced data • Integrated CatBoost and GMM create a powerful model for Pb-Zn prospectivity mapping. • A comprehensive comparison of supervised and semi-supervised has been conducted.
Hajihosseinlou et al. (Wed,) studied this question.