<p>The Bayan Obo deposit is a world-class REE-Fe-Nb (rare earth element-iron-niobium) resource and the most representative carbonatite-type rare earth deposit, holding immense economic and strategic value. However, traditional single geophysical methods have struggled to accurately delineate the deep 3D geometry of its ore bodies, limiting effective resource evaluation and exploration targeting. To clarify the subsurface distribution of the ore-bearing carbonatite, this study proposes an integrated approach that leverages petrophysical properties as a critical link between geological interpretation and multi-geophysical data. Based on a comprehensive dataset of 4019 samples, we systematically analyzed the density and magnetic susceptibility of three major rock types: slate, dolomite, and iron ore. A lithology prediction model was developed using the K-Nearest Neighbors (KNN) algorithm, which was selected after a comparative evaluation against alternatives (e.g., Random Forest, SVM) for its effectiveness in handling small-to-medium datasets and capturing local feature structures. The model achieves high prediction accuracies of 97.7% for ore bodies and 90.9% for wall rocks. Applied to 3D geophysical inversion results, the model reveals that the ore bodies exhibit an east–west strike and dip southward. The deepest ore bodies occur between the main and east mining pits, with the burial depth shallowing toward the eastern and western flanks. Significant ore-hosting carbonatites are also identified in the Dongjielegele area. This work delineates the 3D boundaries of deep-seated ore-hosting carbonatite in the Bayan Obo deposit, while extending the known mineralization depth to 2000 m, effectively translating geophysical anomalies into quantifiable resource potential. The resulting 3D model provides critical insights for formulating future exploration strategies, evaluating resource spatial distribution, and optimizing comprehensive utilization.</p>