Predicting Adsorption Energies on MXene Surfaces Using Machine Learning to Enhance Catalyst Design for the Water–Gas Shift Reaction

Abstract

Efficient prediction of gas-phase adsorption energies on MXene surfaces is critical for advancing materials science applications of these materials. This study integrates data from density functional theory calculations, both in-house and from the literature, to train, validate, and test machine learning models for predicting adsorption energies of various species involved in the water–gas shift reaction (WGSR) mechanism (H₂O, CO₂, H₂, CO, O₂, OH, O, H) on MXenes, considering different compositions and surface terminations (O, H, S, F, Cl, among others). Our database comprises 600 data points with diverse structural, electronic, and adsorption properties. We identify key properties influencing adsorption behavior through data preprocessing and feature selection. Five supervised machine learning models were employed: random forest regression (RFR), XGBoost regression (XGB), artificial neural network (ANN), decision tree regression (DTR), and gradient boosting regression (GBR). Each model underwent cross-validation using 80% of the dataset and testing on a 20% hold-out sample. Results demonstrate that RFR and XGB effectively predict adsorption energies of important species in the WGSR, showing a good correlation between actual and predicted values. Feature importance analysis highlights the significance of crucial adsorbate characteristics, such as the number of radical electrons, the molecular weight, and the total number of valence electrons, as well as key MXene features, such as the Bader charges of the transition metal element (M) and of surface terminations (T), and the standard deviation of block, which refers to the measure of the amount of variation or dispersion of properties among elements within a specific block of the periodic table. This study enhances our understanding of MXene-based materials and provides a promising predictive approach for the adsorption behavior, with implications for catalyst design, such as in the WGSR.