Advanced Copper Prospectivity Mapping in Northwestern India through Machine Learning and Multisource Data Integration

The growing demand for copper, driven by its critical role in green energy technologies such as electric vehicles and renewable energy systems, underscores the need to identify new copper resources. The Aravalli-Delhi Mobile Belt (ADMB), a geologically complex terrain spanning Rajasthan, Haryana, Gujarat, and Delhi, represents significant potential for copper mineralization within its Archaean to Neoproterozoic sequences. In this study, we developed a high-resolution copper prospectivity map for the ADMB by leveraging advanced machine learning techniques and integrating diverse geoscientific datasets. Our methodology incorporated geological features (e.g., proximity to folds, faults, and lineaments), geophysical data (gravity and magnetic anomalies), and remote sensing inputs (SRTM and LANDSAT imagery). Comprehensive processing of potential field data included upward continuation to multiple heights (500 m, 1000 m, 2000 m, 5000 m, 7500 m, 10,000 m, 15,000 m, 25,000 m, and 40,000 m), followed by the computation of first- and second-order directional derivatives, resulting in a total of 154 predictive features. Known copper deposit locations (56 in total) across the ADMB were used as training points, with feature sampling creating the dataset for machine learning model training. We addressed the challenge of class imbalance posed by the limited number of known deposits, by employing synthetic data generation techniques, including Variational Autoencoder (VAE) and Synthetic Minority Oversampling Technique with Generative Adversarial Networks (SMOTE-GAN). Comparative analysis showed that SMOTE-GAN produced more realistic synthetic samples, significantly improving model performance. The enriched datasets were used to train supervised learning models, including Explainable Boosting Machine and Random Forest, optimized within a Positive-Unlabeled (PU) Bagging framework to classify unlabeled regions. Our trained model achieved a predictive accuracy of 95.75% on an unseen dataset. The resulting copper prospectivity map effectively delineates high-probability zones, with nearly all known deposits falling within regions predicted to have probabilities >0.7. Our maps highlight regions of high prospectivity for copper resources that currently lack known deposits, suggesting potential new exploration targets.This demonstrates the robustness of our integrated data approach and machine learning models in identifying unexplored copper-rich areas within the ADMB. Our study highlights the importance of integrating geoscientific data with synthetic data generation to address data scarcity in mineral exploration. The demonstrated scalability of this framework provides a robust solution for prospectivity mapping in other similar Archaean to Neoproterozoic terrains worldwide.