Predicting the HER activity of SACs on MXenes with simple features and interpretable machine learning models

Abstract

The efficient design of single-atom catalysts (SACs) for hydrogen evolution reaction (HER) requires rapid and accurate screening methods capable of handling a large number of candidate materials. Traditional density functional theory (DFT) calculations, while reliable, are computationally expensive, necessitating alternative approaches for high-throughput screening. In this work, we create our own database based on DFT calculations using MXene-based SACs for HER and develop an interpretable machine learning (ML) model that predicts hydrogen adsorption free energy (ΔG_H) with an accuracy of 0.17 eV, relying solely on simple, non-DFT-calculated features. Our approach systematically transitions from a black-box model to a grey-box model, incorporating feature importance analysis to identify key descriptors influencing ΔG_H. This ultimately leads to the development of a glass-box model, where we derive a transparent analytical equation that allows for direct prediction of ΔG_H using raw feature values. Importantly, the final model does not require data standardization or complex computational simulations, making it highly accessible for experimental researchers and enabling rapid, first-order screening of MXene-based SACs for HER.