Abstract: Promoting the advanced recycling and utilization of solid waste resources is crucial for achieving sustainable green building development, with the introduction of nanomaterials emerging as a cutting-edge approach to enhance the overall performance of solid waste concrete. Since porosity is a key determinant of concrete performance and durability, carbon nanotubes (CNTs) can reduce porosity by optimizing microstructure; however, determining their optimal dosage remains challenging under limited experimental conditions. This study investigates pore evolution patterns through compressive strength tests and mercury compression tests. By integrating multi-source data, we developed a weighted XGBoost neighborhood data-enhanced multi-interpolation fusion model based on physical constraints, establishing a porosity prediction model with boundary extrapolation constraints. Results demonstrate that recycled concrete containing 0.102% CNTs exhibits optimal density and performance, with porosity showing a V-shaped trend. SHAP analysis confirmed the nucleating and filling effects of appropriate CNTs concentrations, with the dataset demonstrating excellent fit (R² = 0.8623) and good generalization capability (R_RMSE = 2.0778%). This approach achieves a balance between predictive accuracy and physical interpretability even with small sample sizes.