Abstract
In this work, we present a machine learning-accelerated density functional theory framework for predicting key electrochemical descriptors of oxide perovskites used as oxygen electrodes in solid oxide electrochemical cells, specifically the electronic band gap and oxygen vacancy formation energy. Two classes of models were developed and optimized, tree-based and neural network, using intrinsic properties of the constituent cations as input features. The analyses based on permutation feature importance and SHAP values revealed that both B-site electronic properties and A-site geometric properties play significant roles in determining these descriptors. DFT validation on SrFe0.25Mo0.25Co0.25Mn0.25O3−δ confirmed the machine learning predictions, demonstrating the reliability of the proposed approach. This integrated ML-DFT methodology provides a robust and interpretable framework for accelerating the discovery of functional perovskites for energy conversion applications.