A Machine Learning Perspective on the Brønsted–Evans–Polanyi Relation in Water‐Gas Shift Catalysis on MXenes

Machine learning (ML) has been employed to predict activation energies for key reactions in the water‐gas shift process on pure MXene surfaces, namely, the dissociations of water (H 2 O) and hydroxyl (OH), and the formation reactions of hydrogen (H 2 ) and carbon dioxide (CO 2 ). The dataset includes 92 MXenes, including single‐transition‐metal (M 2 N or M 2 C) and dual‐transition‐metal (M’ 2 M’’C 2 ) compositions, with energies derived from density functional theory calculations from this work or the literature. Several ML models are evaluated, including Random Forest Regressor (RFR), Gradient Boosting Regressor, Artificial Neural Networks, Support Vector Machines, Decision Tree Regressor, and K‐Nearest Neighbors Regressor, identifying RFR as the most accurate. Correlation plots comparing actual versus predicted activation energies further validate the robustness and accuracy of our models. Feature importance analysis reveals that two key factors, the reaction energy and the logarithmic partition coefficient (LogP) of the reactant, are crucial in predicting activation energies. While the former feature aligns well with the Bronsted‐Evans‐Polanyi principle relating to the kinetics and thermodynamics of a given reaction, the latter encompasses different physicochemical properties in only one parameter: polarity, hydrogen bonding, molecular size, molecular shape, and charge. This study underscores the power of ML in catalysis, offering a predictive framework to guide catalyst design.