An advanced spatial statistical framework employing AI approach has been built to model the grade distributions of iron ore deposits in West Singhbhum region of eastern India. The framework treats borehole data as marked point processes, using a Euclidean norm dissimilarity measure to capture spatial proximity. The inherent nonlinearity in the spatial distribution is modeled through deep kernel learning AI methods. To enable effective error reduction in machine learning estimates, features are designed to capture anisotropic spatial variability. As a result, the proposed approach provides a generalized AI framework for nonlinear and anisotropic data, supporting robust grade-tonnage estimation. The model has been applied to iron ore data collected from five different deposits in the West Singhbhum region, covering both contiguous and non-contiguous areas with significant orientation and anisotropy. The results are then compared with simulated data generated by bootstrapping the original dataset after removing spatial anisotropy. This approach offers a robust tool for resource assessment and decision-making in iron ore exploration in the West Singhbhum region.