• Evaluated 10 metaheuristics to optimize K-means for mineral prospectivity. • HOA-KM achieved highest accuracy (AUC: 79.99%, Nd: 6.03) in porphyry modeling. • CSOA-KM and POA-KM ranked second and third in performance metrics. • Confidence Index (CI) improved model reliability and reduced exploration risk. • Results support cost-effective, unsupervised AI-MPM in early-stage exploration. Mineral prospectivity mapping (MPM) conducted within geographic information systems (GIS) frameworks has become a fundamental component of modern mineral exploration. This approach incorporates mineral system-related geospatial data through exploration information systems (EIS). In recent years, artificial intelligence (AI) methods applied to mineral prospectivity mapping (AI-MPM) have emerged as a particularly dynamic and promising research area within the EIS field. Unsupervised AI clustering algorithms offer a solution for reducing uncertainty and increasing accuracy in identifying exploration targets derived from unsupervised AI-based MPM. Moreover, combining metaheuristic algorithms with unsupervised AI clustering enhances the precision and optimization of predictive models. In this study, the K-means clustering, an unsupervised AI clustering algorithms, is employed for unsupervised AI-MPM using geospatial data (9 evidence layers) associated with porphyry copper mineralization (PCM) in the Chahar-gonbad region within the Kerman metallogenic belt, Iran. By integrating 10 top-performing metaheuristic algorithms (as optimization algorithms) with K-means (KM) clustering (e. g., Cuckoo Search Optimization Algorithm (CSOA), Hippopotamus Optimization Algorithm (HOA), Horse Herd Optimization Algorithm (HHOA), Newton-Raphson Optimization Algorithm (NROA), Particle Swarm Optimization Algorithm (PSOA), Poplar Optimization Algorithm (POA), Secretary Bird Optimization Algorithm (SBOA), Tiki-Taka Optimization Algorithm (TTOA), Walrus Optimization Algorithm (WaOA), and Whale Optimization Algorithm (WOA)), the defined classes were optimized. We evaluated and ranked various optimization algorithms based on their impact on K-means clustering performance specifically for porphyry copper potential modeling. The effectiveness of each model in identifying high-potential zones was assessed using Normalized Density (Nd), Success Rate Curve (SRC), and Prediction-Area (P-A plots) methodologies. Our analysis revealed that HOA-KM (Nd for high favorable class: 6.03 and AUC: 79.99% (highest overall)), CSOA-KM (Nd for high favorable class: 5.97 and AUC: 79.65% (second highest among all models) , and POA-KM (Nd for high favorable class: 5.20 (third highest) and AUC: 78.06% (fourth overall) algorithms demonstrated superior performance in optimizing unsupervised AI-MPM, ranking first through third respectively, while the NROA-KM showed minimal effectiveness. The study demonstrates that HOA-optimized K-means clustering represents a particularly efficient approach for regional mineral exploration “Prospecting Stage”, identifying promising areas with reduced exploration risk while minimizing costs through effective utilization of existing exploration datasets. Also, by incorporating the confidence index (CI), this method enhances the reliability of predictions, further minimizing exploration costs through optimal use of existing datasets. The calculation of CI for all 10 optimized unsupervised AI-MPM facilitates more accurate decision-making, ensuring that exploration efforts are focused on the most prospective regions while reducing uncertainty and increasing overall efficiency.
Daviran et al. (2026) studied this question.