Data-driven predictive mapping of gold prospectivity, Baguio district, Philippines: Application of Random Forests algorithm

The Random Forests (RF) algorithm has recently become a fledgling method for data-driven predictive mapping of mineral prospectivity, and so it is instructive to further study its efficacy in this particular field. This study, carried out using Baguio gold district (Philippines), examines (a) the sensitivity of the RF algorithm to different sets of deposit and non-deposit locations as training data and (b) the performance of RF modeling compared to established methods for data-driven predictive mapping of mineral prospectivity. We found that RF modeling with different training sets of deposit/non-deposit locations is stable and reproducible, and it accurately captures the spatial relationships between the predictor variables and the training deposit/non-deposit locations. For data-driven predictive mapping of epithermal Au prospectivity in the Baguio district, we found that (a) the success-rates of RF modeling are superior to those of weights-of-evidence, evidential belief and logistic regression modeling and (b) the prediction-rate of RF modeling is superior to that of weights-of-evidence modeling but approximately equal to those of evidential belief and logistic regression modeling. Therefore, the RF algorithm is potentially much more useful than existing methods that are currently used for data-driven predictive mapping of mineral prospectivity. However, further testing of the method in other areas is needed to fully explore its usefulness in data-driven predictive mapping of mineral prospectivity.