Accurate prediction of thermal conductivity (κ) in doped carbon nanotubes (CNTs) is crucial for advanced thermal management in nanoelectronics. While molecular dynamics (MD) simulations provide physical insights, they are computationally expensive for high-throughput screening. In this study, we propose a Graph Neural Network (GNN) framework to predict the thermal conductivity of nitrogen-doped CNTs under varying strain and doping conditions. A dataset of 5,000 configurations is generated using MD simulations with the AIREBO potential, covering a wide range of chiralities, nitrogen doping concentrations (0–5.%), and tensile strains (0–8%). The GNN model, based on a Message Passing Neural Network (MPNN) architecture, achieves a root mean square error (RMSE) of 0.14 W/mK and a coefficient of determination (R²) of 0.99 on the test set, demonstrating exceptional predictive accuracy. Interpretability analysis via Gradient-weighted Class Activation Mapping (Grad-CAM) reveals that nitrogen atoms and their adjacent strained carbon bonds are the primary phonon scattering centers, significantly reducing κ. Sensitivity analysis confirms that doping concentration and tube diameter are the most influential parameters. The proposed hybrid MD-GNN framework enables rapid and physically interpretable prediction of thermal properties, offering a powerful tool for the design of intelligent thermal materials and strain-based nanosensors.