Targeting porphyry Cu deposits in the Chahargonbad region of Iran: A joint application of deep belief networks and random forest techniques

Abstract

Mineral prospectivity modeling (MPM) is a valid and progressively accepted predictive tool for mapping reproducible potential mineral exploration targets. In this study, a hybrid approach combining unsupervised deep belief networks with supervised random forest (DBN-RF) is performed to delineate potential exploration targets for porphyry Cu deposits in the Chahargonbad region of Iran. Firstly, a mineral system model for porphyry Cu deposits is established, and relevant targeting criteria are delineated based on comprehensive exploration datasets. Subsequently, within this hybrid framework, the DBN extracts deep implicit feature information, which is then utilized as input for the RF. The comparative results on the performance of the hybrid model and the RF model trained by the primary targeting criteria, in terms of the improved prediction-area plot, demonstrate that the DBN-RF prospectivity model outperformed the RF-generated model with an overall efficiency of 0.53. This hybrid model accurately identified 81.97 % of known Cu deposits within an investigation area of 18.03 %, with primary trends aligned with the primary faults and volcanic units of the region. This study demonstrates effective performance of DBN-RF in identifying exploration targets for porphyry Cu deposits at regional scale and also highlights the potential of deep learning-based methods for successful MPM.