Uncertainty quantification in genetic algorithm-optimized artificial intelligence-based mineral prospectivity models: automated hyperparameter tuning for support vector machines and random forest

This study investigates the challenges and opportunities presented by integrating genetic algorithm (GA) with artificial intelligence-based mineral prospectivity mapping (AI-MPM) for porphyry copper exploration. We develop a systematic approach to address the uncertainties arising from the use of multiple algorithms in hybrid models. The research focuses on the Kerman belt in Iran, a region known for porphyry-type Cu mineralization. Our methodology employs four AI algorithms including support vector machines (SVM), random forest (RF), genetic-SVM (GSVM), and genetic-RF (GRF) to create predictive models and assess uncertainties. We utilize an average voting procedure and C-A fractal classification to quantify uncertainties in target criteria and evaluate the effectiveness of both standard (SVM and RF) and GA-optimized machine learning algorithms (GSVM and GRF) in enhancing mineral prospectivity models. The study aims to: (i) develop accurate predictive AI-based models for porphyry copper mineralization, (ii) quantify uncertainties in target identification, (iii) assess the performance of GA-optimized AI-MPM in well-researched and underexplored areas, (iv) identify low-risk exploration targets with high potential. Our results demonstrate the utility of SVM and RF in mineral exploration, with the GA-optimized AI-MPM approach successfully identifying low-risk zones covering 12% of the study area. These zones are predicted to contain 74% of the known mineral deposits, highlighting the method's effectiveness in focusing exploration efforts. This research contributes to the advancement of AI-MPM techniques by addressing uncertainty quantification and optimization, potentially improving the efficiency and success rate of mineral exploration projects.