This approach achieves explainability for ML models through post-hoc SHAP analysis and Symbolic Regression. By merging ML techniques with XAI, we bridge the gap between data-driven methods and domain-specific knowledge, aiming to elucidate the intricate relationship between input features and adsorption energy.
-
Post-hoc Analysis: Utilizes shallow machine learning techniques alongside post-hoc XAI analysis using the SHAP library. Post-hoc analysis reveals feature importance, inter-feature correlation, and and elucidates the influence of feature values on predicting adsorption energy values.
-
Symbolic Regression : Employs Symbolic Regression (SR) to derive mathematical equations for computing adsorption energy from input features. These equations are compared to those obtained from traditional theory or experimental-based methods, highlighting their potential complementarity.