A New Frontier in Mineral Exploration: Hybrid Machine Learning and Bat Metaheuristic Algorithm for Cu-Au Mineral Prospecting in Sonajil area, E-Azerbaijan

Abstract

Abstract Modeling mineral prospectivity is a challenging strategy in characterization of blind ore zones and mineral prospecting. Therefore, the application of advanced spatial modeling techniques and machine learning algorithms is important in exploration pattern recognition. In this study, a combination of a support vector machine (SVM) and the bat algorithm (BA) has been employed to improve the classification and produce an optimal prospectivity map of porphyry copper-gold deposit in the Sonajil area, NW of Iran. In the proposed model, BA was used to optimize the SVM parameters. In data preparation, evidence layers of 10 exploratory features (geological, geochemical, remote sensing and boreholes) were prepared. Then based on 100-meter buffering around boreholes, the data (training and testing) was selected. For mapping mineral prospectivity in the area, the results of two predictive single SVM and hybrid SVM-BA models were compared. Both models were trained by way of predictor maps and then the models performance were evaluated by confusion matrix and receiver operating characteristic (ROC) curve. The results showed that both predictive models had an obvious optimization effect on recognizing the Cu-Au mineralization pattern but the hybrid model had higher accuracy in anomalous zone enhancement, with ROC values more than 0.8, indicating this optimization was successful and the selected optimal model is the best predictor for mineral prospectivity in the area. The delineated targets are also in accordance with the characteristics of the area metallogenic system showing that the established hybrid model is an effective tool in mineral prospectivity mapping.