Elucidating the key drivers of twin formation in copper alloys via a small-sample AutoML framework

Abstract The twin structure in materials has become a current research hotspot. To clarify the intrinsic effect of alloying elements on twin formation in copper, this study constructed descriptors based on 31 sets of first-principles calculation data and performed a three-level screening on the 132 generated descriptors, including constant elimination, correlation filtering, and comprehensive importance ranking. Six key features were identified through forward accumulation cross-validation. Automated machine learning(AtuoML) was used for modeling, tuning, and validation, successfully distinguishing the tendency of elements to form twins and mapping the results onto the entire periodic table. SHAP interpretative analysis shows that the minimum covalent radius and maximum GSbandgap play a dominant role in twin formation. This method not only explains the mechanism of twin formation at the atomic scale but also provides a basis for the rapid screening of alloy elements.