Geochemical modeling of orogenic gold deposit using PCANN hybrid method in the Alut, Kurdistan province, Iran

In this paper stream sediments based geochemical exploration program with the aim of delineating potentially promising areas by a comprehensive stepwise optimization approach from univariate statistics, PCA, ANN, and fusion method PCANN were under taken for an orogenic gold deposit located in the Alut, Kurdistan province, NW of Iran. At first the data were preprocessed and then PCA were applied to determine the maximum variability directions of elements in the area. Subsequently the artificial neural network (ANN) was used for quick estimation of elemental concentration, as well as discriminating anomalous populations and intelligent determination of internal structure among the data. However, both the methods revealed constraints for modeling. To overcome the deficiency and shortcoming of each individual method a new methodology is presented by integration of both “PCA & ANN” referred as PCANN method. For integrating purpose, the detected PCs pertinent to ore mineralization selected and intruded to neural network structure, as a result different MLPs with various algorithms and structures were produced. The resulting PCANN maps suggest that the gold mineralization and its pathfinder elements (Au, Mo, W, Bi, Sb, Cu, Pb, Ag & As) are associated with metamorphic host rocks intruded by granite bodies in the Alut area. In addition, more concealed and distinct Au anomalies with higher intensity were detected, confirming the privileges of the method in evaluating susceptibility of the area in delineating new hidden potential zones. The proposed method demonstrates simpler network architecture, easy computational implementation, faster training speed, as well as no need to consider any primary assumption about the behavior of data and their probability distribution type, with more satisfactory predicting performance for generating gold potential map of the area. Comparing the results of three methods (PCA, ANN and PCANN), representing the higher efficiency and more reliability of PCANN with lesser training time, simple structure, and correlate components while avoiding the duplicate entry of data to network. This study also suggests that in many similar cases integrated methods have capability to fix bugs more effectively and successfully in exploration programs.