<p>The demand for metals and the potential future supply deficit of certain metals have influenced the search for new deposits, even in exploration–immature terranes. With the availability of pre-competitive regional geoscientific datasets, the computer-based mineral prospectivity mapping (MPM) approach has become the hinge of modern-day mineral exploration programs. However, the execution of MPM requires a selection and translation of mappable proxies based on well-established mineral systems. In this study, we collated a mineral systems framework and selected targeting evidential proxies to target orogenic gold mineralization within one of the exploration–immature gold terranes, the Pietersburg Greenstone Belt (PGB), South Africa. We also adopted and compared data-driven random forests (RF) and knowledge-driven fuzzy inference models to test the collated framework and the selected evidential proxies in generating prospectivity maps. Both models showed acceptable predictive capabilities based on receiver operator characteristic (ROC), success-rate and improved prediction–area (P–A) plots. However, the RF model outperformed the fuzzy inference model based on the ROC measure with an area under curve of 0.949 compared to 0.806 for the latter. The improved P–A plot also showed that the RF model was the better model based on an overall performance indicator (<i>O</i>_<i>e</i><i> = </i>true-positive rate (<i>TP</i>_<i>r</i>)<i>—</i>false-positive rate (<i>FP</i>_<i>r</i>)) with 0.42 compared to 0.17 for the fuzzy inference model. The predictivity capabilities of both models indicated that the proposed framework used to select the evidential proxies was sensitive to delineating gold mineralization within the PGB.</p>