Machine learning prediction of metallic glass forming ability: The pivotal role of relative energy

• Involving relative energy (RE) into ML models boosts GFA prediction accuracy. • SHAP analysis shows RE’s vital role and its correlation with GFA of BMGs . • The proposed framework defines key parameter ranges, guiding for better GFA. • Future work should focus on evaluating RE's interactions for broad MG systems. Despite the copious prior investigations into the forming ability of bulk metallic glasses (BMGs), accurately predicting glass forming ability (GFA) has persisted as a formidable challenge. By incorporating relative energy (RE, defined as the difference between the total energy of the alloy and the reference state of its constituent elements), which has been hitherto largely overlooked in machine learning (ML) prediction analyses, we demonstrate effective improvements in multiple ML models involving Extreme Gradient Boosting, Support Vector Regression, Linear Regression, and Decision Trees. Moreover, feature importance analysis based on SHAP (SHapley Additive exPlanations) summary plot indicates that RE ranks first in all four ML models, highlighting its crucial role in ML prediction of GFA, providing a new perspective for understanding and predicting GFA of MGs.