Evaluation of machine learning algorithms for spatial predictive modeling of Au prospectivity

Definition of the efficient ore-forming processes which are considered as mineralization controls is a fundamental stage in mineral prospectivity mapping. In this contribution, five targeting criteria of geochemical, geophysical, geological, structural and hydrothemal alteration data related to epithermal and Carlin-type Au deposits in Takhte-soleyman district, NW Iran, were integrated. For creation of multi-element geochemical layer, principal component analysis was firstly conducted on stream sediment data of 8 selected elements and it was found that PC2 is the representative of Au-As-Sb elemental association in the study area. Then, 10 fuzzified efficient evidence layers were selected based on area under the curve (AUC) of success-rate curves and prepared for generation of predictive models of Au mineralization. For this purpose, two supervised machine learning algorithms, namely multi-layer perceptron (MLP) neural network and support vector machine (SVM) with RBF kernel were used. Comparison of the generated models demonstrates that the latter is more succeeded in delineating exploration targets than the former one.