<p>Mineral exploration increasingly depends on advanced machine learning techniques to identify concealed deposits in complex geological environments. Traditional 3D mineral prospectivity modeling (3D MPM) often faces challenges with suboptimal feature prioritization in multi-dimensional data, resulting in inefficient targeting and increased exploration risk. This study introduces a novel 3D MPM approach that integrates Squeeze-and-Excitation Networks (SE Networks) with Convolutional Neural Networks (CNNs) to map gold mineralization in the Siahcheshmeh intrusion-related gold deposit in northwest Iran. Detailed 3D models were constructed using geological, geochemical, and geophysical datasets, including magnetic and induced polarization surveys, to identify high-potential zones. The CNN + SE model outperformed standard CNN, Support Vector Machines (SVM), and Logistic Regression (LR), achieving an accuracy of 98.9% (compared to 97.3%, 89.2%, and 88.9%, respectively). Validated with drillhole data, the model accurately delineated prospective zones and demonstrated robustness in altered terrains. By enhancing feature selection and spatial interpretability, this approach significantly improves exploration efficiency, reduces the target volume, and facilitates risk-based decision-making in environments with concealed deposits.</p>