Advances in artificial intelligence and data science are creating new opportunities to improve geological domain definition, a critical step in mineral resource modeling and uncertainty management. This paper presents a machine learning (ML)-based workflow for multivariate domain definition, combined with geological judgment, applied at an underground gold deposit, demonstrating how ML can be integrated with geological constraints to deliver domains for robust resource estimation and business decision‑making.
The workflow incorporates a Dual Space Ensemble Spatial Clustering algorithm to group samples based on both spatial proximity and multivariate similarity, addressing limitations of traditional manually defined domaining strategies. Continuous variables are jointly analyzed to capture complex multivariate relationships that reflect underlying mineralization controls. Domain boundaries are then modeled using Indicator Kriging and probability thresholding calibrated to nearest-neighbor volumes to ensure global unbiasedness. Large‑scale structural and lithological boundaries are used as constraints, allowing the algorithm to operate within geologically meaningful frameworks rather than relying solely on statistical partitions. The ML‑based proposed domains were reviewed in collaboration with the lithological, structural and deposit knowledge from the geology modeling team and subsequently used as the basis for the final estimation domains definition, resulting in a significant change to the domaining strategy.
The final estimation domains exhibit improved statistical separation while maintaining spatial continuity and honoring key geological features. Statistical analysis, contact plots, and multivariate scatter plots demonstrate clearer domain separation compared with conventional approaches, with reduced mixing across boundaries. Importantly, the workflow provides a transparent basis for domain definition, enabling practitioners to better understand and communicate the uncertainty associated with geological interpretation and domain geometry.
This work highlights how advanced ML methods can enhance resource modeling by supporting more objective, data‑driven decisions while remaining firmly anchored in geological knowledge. When embedded within appropriate governance and validation frameworks, such approaches offer significant potential to improve confidence in domain boundaries and the resource estimate, support mineral resource and reserve reporting, and unlock new opportunities for value creation in complex deposits.