https://youtu.be/dcg8tugMYoU

Mineral exploration targeting relies heavily on accurately identifying promising areas, often using supervised machine learning algorithms like Random Forest. A significant challenge in this process is the precise selection of non-deposit locations, which are essential for training accurate predictive models but can be difficult to determine reliably. Incorrectly choosing these points can lower the prediction rate and introduce systematic bias. To tackle this problem, a study applied deep autoencoders (DAE)—a type of neural network designed to compress and reconstruct data—for unsupervised labeling to help identify suitable non-deposit locations within recognized non-prospective regions in the Varzaghan district, NW Iran. This area is considered highly prospective for porphyry copper deposits. The researchers compared this innovative DAE-based approach for selecting non-deposit points with a method relying on expert geological knowledge. Both sets of non-deposit locations, along with known mineral occurrences, were used to train Random Forest models for porphyry copper prospectivity mapping. The models incorporated various evidence layers crucial for identifying potential deposits, including multi-element geochemical signatures derived using techniques like Staged Factor Analysis (SFA) and the Geochemical Mineralization Probability Index (GMPI), fault density, and proximity to alteration zones (argillic and phyllic) and intrusive rocks. The evaluation results, using methods such as the prediction-area (P-A) plot and normalized density index (Nd), showed that the DAE-enhanced Random Forest model achieved a higher prediction rate and better correlation with key geological indicators like intrusive rocks compared to the model based on expert knowledge. This demonstrates that integrating unsupervised labeling via DAE with supervised algorithms offers a robust and effective method for mineral prospectivity mapping, holding significant potential for identifying new exploration targets in similar mineral systems. P. Geo. Ricardo A Valls, M. Sc. and Geo Gadfly Valls Geoconsultant ORCID ID- https://orcid.org/0000-0002-5421-0914 Scopus Author ID: 7003369619/35335510700 ResearcherID: S-6604-2018 If you like this content, please "buy me a coffee" https://www.buymeacoffee.com/goldendroplets