<p>Accurate and efficient prediction of the mechanical properties of Magnesium (Mg) alloys is crucial for reducing experimental costs and screening ideal materials. This study proposes an interpretable machine learning framework combining data augmentation via the multi-objective particle swarm optimization (MOPSO) algorithm and hardness (HV) assistance. By integrating virtual samples generated by MOPSO with 271 sets of original experimental data to train the XGBoost model, it is found that the model performance improved significantly after data augmentation. After incorporating HV into the input and optimizing hyperparameters, the generalization ability of the XGBoost model was significantly enhanced, which confirms that HV is a core surrogate parameter capable of capturing detailed microstructural characteristics. The coefficients of determination (R²) on the test set reached 0.99 (for yield strength, YS), 0.99 (for ultimate tensile strength, UTS), and 0.98 (for elongation, EL), respectively. Shapley Additive Explanations analysis verifies that HV is a key determinant of the model performance, while accurately identifying the key regulatory thresholds, including the action thresholds of alloying elements, processing parameters and heat treatment parameters, thus clarifying the synergistic regulation mechanism of various factors on the strength and toughness of Mg alloys. Experimental validation using six alloy combinations shows that the model’s prediction errors for unknown alloys were all &lt; 10%. This study establishes a high-precision and interpretable prediction paradigm for the mechanical properties of Mg alloys, providing clear guidance for alloy composition and process optimization.</p>