Predictive modeling of orogenic gold prospectivity in the Wawa area, western Nigeria, using three most commonly-used machine learning algorithms

The unavailability of prospectivity maps for orogenic gold deposits in western Nigeria has been a significant setback in the federal government's efforts to attract mining investors. Prospectivity modeling is a potent exploration tool that allows for the segregation of large expanses of land into high and low prospectivity zones. To support the Nigerian government efforts in its diversification of the Nigerian economy towards the solid mineral sector and attract mining investments into the country, this research focused on the use of machine learning (ML) models for orogenic gold prospectivity modeling in the Wawa area, a southern extension of the gold-bearing Zuru Schist Belt of western Nigeria. A mineral system approach was applied in this study to translate the characteristics of orogenic gold deposits into nine predictor maps, accounting for heat source, fluid pathway, trapping mechanisms, and favorable host rocks needed for the formation of such deposits. The ML algorithms considered in this research were artificial neural network (ANN), support vector machine (SVM), and random forest (RF). Each of these predictive models were trained using nine predictor maps and 174 data points, consisting of known gold deposit and non-deposit points. The RF predictive model had the highest prediction efficiencies with Accuracy, AUC, and Kappa index values of 0.91, 0.96, and 0.83, respectively; the SVM model gave similar performance metrics such as accuracy (0.89), recall rate (0.88), F1 score (0.88), and Kappa index (0.77) as the ANN, but had a higher precision value (0.91) and lower AUC value (0.91) than the ANN (0.89 and 0.92, respectively). The results suggest that the three ML predictive models are suitable for orogenic gold prospectivity modeling in the study area. Further exploration for orogenic gold deposits in the Wawa area will likely be successful within the delineated high prospectivity zones. • Sufficient geo-exploration datasets exist for predictive modeling of orogenic gold deposits in western Nigeria. • Gold mineralization in the Wawa area, western Nigeria, are mostly controlled by regional faults within the area. • Random forest and neural networks performed best in predicting orogenic gold deposits in Wawa area, western Nigeria.