Por: Alejandro Bueno y Fernando Saez, SRK Consulting Perú y Antoine Caté, SRK Consulting Canadá.

Abstrac

This study evaluates the application of Graph Convolutional Networks (GCN) for mineral prospectivity mapping (MPM) of copper porphyry-skarn in the Andahuaylas-Yauri belt, a significant metallogenic zone in southern Peru.

Publicly available geoscientific data, including geochemistry, lithology, and lithological contacts, were integrated into evidence maps and used to train four machine learning models: Random Forest (RF), Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Graph Convolutional Network (GCN). Spatial cross-validation using k-means clustering was employed to mitigate spatial bias arising from the limited number of known mineral occurrences. 

The results show that MLP and GCN achieved the highest AUC values (0.910 and 0.907, respectively), outperforming RF and CNN. However, GCN predictions exhibited greater spatial and geological coherence, identifying favorable areas consistent with the expected mineralization processes described by the mineral systems model. 

Graph neural networks effectively incorporate the spatial structure of the data, improving the delineation of priority areas for copper exploration. This approach represents a significant advance in integrating artificial intelligence into mineral exploration.