A comparative analysis of artificial neural network (ANN), wavelet neural network (WNN), and support vector machine (SVM) data-driven models to mineral potential mapping for copper mineralizations in the Shahr-e-Babak region, Kerman, Iran

Abstract

The study area located in the southern section of the Central Iranian volcano–sedimentary complex contains a large number of mineral deposits and occurrences which is currently facing a shortage of resources. Consequently, prospecting potential areas in the deeper and peripheral spaces has become a high priority in this region. Different direct and indirect methods try to predict promising areas for future explorations which most of them are very time-consuming and costly. Therefore, applying an efficient method which can model the mineral potential areas correctly and decrease the uncertainty is necessary. The main objective of this study is to predict new explanatory areas associated with Cu mineralization in Shahr-e-Babak mineral region using three data-driven models, namely artificial neural network (ANN), wavelet neural network (WNN), and support vector machine (SVM). Most of the researches show that the capability (i.e., classification, pattern matching, optimization, and prediction) of an ANN is suitable for inheriting uncertainties and imperfections found in mining engineering problems considering its successful application. Despite all of the applications of ANNs in mineral potential mapping, they have limitations. In this paper, an alternative method of mineral potential mapping is presented which is based on integration between wavelet theory and ANN or WNN. The obtained results by WNN are in good agreement with the known deposits, indicating that WNN method with POLYWOG 3 transfer function have high complex ability to learn and track unknown/undefined complex systems. The efficacy of this type of network in function learning and estimation is compared with ANNs. The simulation results indicate a decrease in estimation error values that depicts its ability to enhance the function approximation capability and consequently exhibits excellent learning ability compared to the conventional neural network with sigmoid or other activation functions. Also, for better comparison, the results of neural network and wavelet neural network were evaluated using support vector machine.