Abstract: The Kangjiawan large-scale Pb-Zn deposit in central Hunan exhibits significant deep exploration potential, necessitating targeted studies on deep ore body localization and prospectivity prediction. To address this challenge, a three-dimensional (3D) implicit modeling approach integrated with optimized machine learning algorithms was employed to predict deep-seated ore targets within the deposit, yielding highly promising results. The methodological workflow is as follows: First, drilling data and geochemical data were integrated to construct a standardized geological model dataset. Then, a 3D geological-geochemical model of the deposit was established based on the Leapfrog Geo implicit modeling platform, revealing the spatial distribution patterns of strata, ore bodies and ore-forming elements. Finally, the performance differences and prediction effects of three models, namely Random Forest (RF), Back Propagation Neural Network (BPNN) and Extreme Gradient Boosting (XGBoost), were compared systematically. The results demonstrate that the Random Forest model exhibits the optimal comprehensive performance, while the Back Propagation Neural Network model performs slightly inferior to the former. Their predictions show remarkable spatial consistency, and the inferred geometries of potential ore bodies align closely with established geological controls and mineralization patterns. Through the comprehensive analysis of feature importance of the RF model, verification of prediction results, and integration of regional geological laws, this study reveals that the regional large-scale thrust fault F22, together with favorable host strata—including the Gaojiatian Formation (Jurassic), and the Douling and Qixia Formations (Permian)—as well as silicified alteration zones, constitute the primary structural and lithological controls on mineralization. The spatial distributions of sulfur (S) and gold (Au) exhibit strong positive coupling and high mineralization intensity. Based on ensemble predictions from the RF and BPNN models, two high-potential exploration targets have been delineated. Both targets conform to the recognized mineralization trend at Kangjiawan: ore bodies extend southward along interlayered silicified fracture zones and are preferentially hosted within the Permian Qixia and Dangchong Formations. These results provide a robust geoscientific foundation for guiding deep-level exploration at the Kangjiawan deposit.