Data-driven glass-forming ability for Fe-based amorphous alloys

Fe-based amorphous alloys have garnered significant academic interest due to their distinctive properties and promising application prospects. However, their limited glass forming ability (GFA) presents a significant challenge in engineering applications. This study utilizes three machine learning techniques, namely artificial neural networks (ANN), Random Forest (RF), and eXtreme Gradient Boosting (XGB) to predict the maximum amorphous diameter (Dmax) of Fe-based amorphous alloys based solely on chemical composition. Three oversampling techniques were implemented to tackle the uneven distribution in the original dataset, significantly enhancing model performance. Additionally, Shapley Additive Explanation (SHAP) was employed to interpret the constructed models, revealing pertinent rules regarding GFA. The applied techniques and revealed rules contribute to a statistical comprehension of GFA, potentially aiding in the discovery of novel Fe-based amorphous alloys.