<p>This research introduces an unsupervised machine learning framework for mineral prospectivity mapping (MPM), integrating well-established clustering algorithms, including K-means, K-medoids, and Self-Organizing Maps (SOM), with a customized data preprocessing strategy aimed at improving copper (Cu) prospectivity analysis in the underexplored Kerman metallogenic belt in southeastern Iran. The innovation lies in the development of a hybrid methodology that combines multifractal inverse distance weighting (MIDW), factor analysis (FA), and fuzzy logic-based standardization to effectively extract and normalize complex geoscientific patterns from diverse evidence layers, including geochemical surveys, airborne magnetic data, remote sensing imagery, and structural features. This approach enables the detection of mineralization-related anomalies without relying on training samples, thus overcoming key limitations of conventional clustering techniques in mineral exploration. Clustering was applied to a large-scale dataset containing 66,983 spatial records and 12 features. A normalized density index was used to evaluate and compare the spatial coherence and predictive strength of the resulting prospectivity models. Among the tested algorithms, K-means showed superior performance in identifying copper-rich zones. The proposed methodology significantly improves the accuracy and relevance of unsupervised AI models for MPM, offering valuable insights for targeted exploration and sustainable resource development in the Kerman region.</p>