Integrating multi-source geoscience data with RF-weighted conditional variational autoencoders (CVAE) for porphyry copper prospectivity mapping: an application to the Ardestan district, Central Iran

Porphyry Cu deposits in the Urmia–Dokhtar magmatic belt constitute important exploration targets, yet the heterogeneous and multi-scale nature of geological, geochemical, and remote sensing data introduces substantial uncertainty in prospectivity mapping. This study develops a data-driven framework for porphyry Cu prospectivity assessment in the Ardestan 1:100,000 sheet through the integration of multi-source datasets and deep learning. Structural, lithological, alteration, and stream-sediment geochemical information were transformed into fuzzy membership layers and objectively weighted using the Random Forest algorithm. These weighted layers were subsequently analyzed using a conditional variational autoencoder (CVAE), where reconstruction error was used as an indicator of mineralization prospectivity. The resulting prospectivity map delineates several coherent high-potential zones that show strong spatial correspondence with major fracture systems, hydrothermal alteration halos, and Cu geochemical anomalies. Five priority targets were identified, each characterized by distinctive structural–hydrothermal signatures indicative of porphyry-style mineral systems. To address data scarcity challenges, the workflow incorporates a robust validation framework comprising TSS, MCC, and IoU metrics, alongside an OAT sensitivity analysis that confirms the spatial stability and geometric integrity of the identified anomalies under weight perturbations. External validation using independent field-based reference indices demonstrates that the proposed framework achieves superior predictive performance compared to a conventional CRITIC–WLC integration model. By employing the Random Forest algorithm as a surrogate explanatory model within the XAI framework, the workflow bridges the gap between predictive accuracy and geological interpretability. These findings indicate that the proposed Fuzzy–RF–CVAE workflow effectively reduces exploration uncertainty and provides a transferable, generalizable strategy for data-driven mineral prospectivity mapping in Ardestan and comparable magmatic arc environments.