Machine Learning Aided Prediction of Glass-Forming Ability of Metallic Glass

The prediction of the glass-forming ability (GFA) metallic glasses (MGs) can accelerate efficiency their development. In this paper, a dataset was constructed using experimental data collected from literature and books, machine learning-based predictive model established to predict GFA. Firstly, classification based on size critical diameter (Dmax) determine whether an alloy system could form glass state, with accuracy rating 0.98. Then, regression models were crystallization temperature (Tx), transition (Tg), liquidus (Tl) MGs. R2 obtained in test set greater than 0.89, which showed that had good accuracy. key features used by analyzed variance, correlation, embedding, recursive, exhaustive methods select most important features. Furthermore, improve interpretability model, feature importance, partial dependence plot (PDP), individual conditional expectation (ICE) for visualization analysis, demonstrating how affect target variables. Finally, taking Zr-Cu-Ni-Al MGs as example, genetic algorithm optimize composition high GFA compositional space, achieving optimal design composition.