Data-driven mineral prospectivity mapping (MPM) with random forests (RF) has been documented in various brownfield zones hosting known mineral deposits of the type being sought. However, the practical implementation of RF-based MPM is crippled by uncertainties stemming from the exploration targeting criteria used as predictors for this algorithm; some exploration targeting criteria are poorly linked to known mineral deposits. Aiming to modulate the effects of this type of uncertainty in RF-based MPM, this study employs three parameters, namely the Gini impurity index, least-squared errors, and receiver operating characteristics curves, for sensitivity analysis of exploration targeting criteria. These were used to assess whether the exclusion or inclusion of specific targeting criteria can improve the results of RF-based MPM. Sensitivity analysis was applied to a set of mineral systems-derived exploration targeting criteria in the Ahar-Arasbaran area, located in the Alborz–Azarbaijan magmatic belt (AAMB) of northern Iran. It is demonstrated here that the exploration targeting criteria that are poorly linked to known deposit sites deteriorate the performance of RF-based MPM, the exclusion of which can significantly improve the performance of MPM. The sensitivity analysis adopted in this paper reduces the search space for complimentary surveys and improves the correlation of models generated with known mineralized sites; consequently, one may achieve an improved performance of data-driven MPM by applying the strategy employed in this study.
