Activating non-basal <c + a> slip is essential to enhance the ductility and formability of Mg alloys at room temperature. To determine favorable dopant species for lowering the energy barrier of (10 \(\overline{1 }\) 1) slip system in Mg alloys, we trained various machine learning models using a dataset from 106 distinct elemental species and 29 unique calculations obtained from density functional theory (DFT). The models, including regression decision trees, random forest, and linear and polynomial regression, prioritized features affecting the unstable stacking fault energy (USFE) of the slip system, crucial for activating non-basal <c + a> slip. A third-degree polynomial regression model was selected for its acceptable accuracy without significant overfitting, enhanced by Shapley values to determine the most significant features affecting the USFE. The forces on Mg atoms adjacent to the dopant, the angle formed by adjacent Mg in relation to the dopant atom, and localized charges on the adjacent atoms to the dopant emerged as significant for determining the USFE. The predictive capability of the model was validated using lasso regression, showing accurate prediction of the USFE values for ternary Mg alloys. Our model provides a strong baseline system for determining successful alloying combinations to potentially enhance Mg ductility at room temperature.
