Oral: Explainable machine learning to predict the properties of ABX₃ perovskite materials for photocatalytic applications

The positions of electronic band edges are crucial for assessing a material's ability to function in solar energy conversion devices that generate fuels from sunlight. In this study, we trained and compared explainable machine learning (ML) algorithms for predicting the band edge position of the conduction band minimum (E_CBM) and valence band maximum (E_VBM) of perovskite materials that have the formula ABX₃ with the A, B, and X referring to the 3 elements that make the cubic 3-dimensional framework of the perovskite compounds. Six supervised learning models: CatBoost, Random Forest, XGBoost, LightGBM, SVM, and Decision Tree were employed to study the band edge position of these materials, and the relationships that exist between the input features in predicting these materials are discussed. Input features influencing the ML model decisions were initially determined by performing a correlation analysis on the multi-dimensional input feature space. This shows features with high collinearity and features with limited correlation. The ML models that were trained on the screened input features show that CatBoost and XGBoost models yielded the least predictive errors and the highest coefficient of determination of R² ≥ 91% than other approaches in the testing phase for E_CBM. In addition, for E_VBM, CatBoost, XGBoost, and Random Forest performed better than other models in the testing phase with R² value of approximately 99%. Furthermore, the Shapley Additive Explanation (SHAP) was used to explain the models based on the elemental composition of each perovskite compound and the features used from the physics standpoint. One key insight gained from the SHAP analysis is that the Pauling electronegativity of the B site cation contributes significantly to predicting E_CBM, while that of the X site anion plays a significant role in E_VBM prediction. This further confirms that these sites of the cubic perovskites play an important role in the electronic properties of this class of materials and reveal the potential of ML to predict properties useful in photocatalytic applications.