Abstract: The Zhaishang-Liba mining district in the West Qinling Orogen is one of the key exploration areas for the National Strategic Action for Mineral Exploration Breakthrough in China. Due to the thick shallow surface cover, traditional exploration methods face severe challenges. Mineral prospectivity mapping (MPM) driven by big data and artificial intelligence (AI) provides a new perspective for identifying weak mineralization anomalies in covered terrains. However, during the generation of positive samples, existing models rarely consider the spatial heterogeneity of deposit scales and metallogenic energy fields, which restricts the accuracy and reliability of the predictive models. Therefore, this paper proposes an innovative positive sample construction method dynamically constrained by the "resource-area" relationship. Integrating multi-source exploration data including geological, geochemical, and remote sensing data, 17 predictor variables were extracted to construct a Random Forest (RF) predictive model, coupled with the SHAP (SHapley Additive exPlanations) method for geological interpretability analysis. The predictive model demonstrates excellent generalization and classification performance. Under a probability threshold of 0.52, it successfully captures 92% of the known gold deposits within merely 8% of the predicted area. SHAP analysis reveals that variables such as Au, Sn, Cu, Zn, fault density, Sb, As, and proximity to intrusions play dominant roles in the prediction. The predicted high-probability areas spatially exhibit a "linear" distribution pattern significantly controlled by the superposition of multi-stage faults, and an "annular" pattern surrounding the "Five Golden Flowers" composite intrusions, profoundly reflecting the metallogenic dynamic mechanism that gold mineralization in this area is strictly controlled by the spatiotemporal coupling of the "tectonic-magmatic" system. Based on the prediction results, 6 deep exploration targets were delineated. Notably, target P1 was verified by a deep drill hole (ZK4-1), which successfully intercepted a gold mineralization body with a maximum grade of 1.24 g/t. The study demonstrates that the intelligent mineral prospectivity mapping framework constrained by deposit scale holds significant scientific indicative value and application prospects in covered areas.