<p>Machine-learning (ML) approaches are gaining traction throughout the mining sector to improve modeling processes, particularly in the development of geometallurgical models, which integrate geological, mineralogical, and metallurgical information to predict processing performance across an orebody. There are many unanswered questions about the validity of ML-generated domains compared to actual geological and process characteristics. A case study of the Escondida porphyry Cu deposit is used to develop an ML-based methodology for generating, analyzing, and validating geometallurgical domains (spatially distinct regions of an orebody characterized by similar geological and metallurgical response). We applied and evaluated three clustering models for domain estimation at Escondida: K-means, Hierarchical Clustering, and Gaussian Mixture Models (GMM). The methodology incorporates Principal Component Analysis (PCA) as a dimensionality-reduction step, transforming a large set of variables into a smaller set. Model performance is evaluated based on three metrics: Silhouette Score, Calinski-Harabasz Index, and Davies-Bouldin Index. These metrics showed K-means performed best, with the optimal number of clusters being k = 3, and clustering behavior showing consistency across varying dataset conditions. K-means robustness was examined against variations in dataset size, sampling balance, and cluster dominance. The resulting clusters delineate spatial patterns consistent with the geometallurgical domains previously defined at Escondida, particularly with respect to hardness, mineralogy, and recovery zones. This study demonstrates that a data-driven ML approach can effectively capture and represent geometallurgical features, offering new insights and serving as a complementary tool for geometallurgical domain generation.</p>