التكهن بتابع العمل لمواد ثنائية الأبعاد باستخدام تعلم آلة قابل للتفسير

Abstract

In this study, we designed two machine learning models to predict the work functions of 2D materials using composition-based features. We used RF and XGBoost algorithms and trained them on 4000 materials from the Computational 2D Materials Database (C2DB), achieving very good results on the test set:  and  , with XGBoost slightly outperforming RF. To interpret the results, we used Permutation Feature Importance (PFI) and SHAP analyses. As indicated by PFI, the main features contributing to performance are average electronegativity, column number in the periodic table, and covalent radius. SHAP explains how each feature affects performance; it shows that increases in average electronegativity and column span lead to higher work-function values, which aligns with physical intuition. The models provide acceptable predictions and interpretable design guidance, allowing targeted research on 2D materials for electronic devices.