Adsorption Energy Prediction Model for CO2 Reduction on Electrocatalysts Containing Previously Unencountered Metal Elements.

Electrochemical carbon dioxide reduction (CO2RR) using electrocatalysts has gained attention for its potential to convert atmospheric CO2 into value-added chemicals. Recently, machine learning (ML) has emerged as a promising approach for catalyst development. In previous studies, catalyst screening using ML performed using adsorption energy databases obtained from density functional theory calculations has yielded successful results. However, the full potential of ML in catalyst discovery for rapid screening has been hindered by the limitation of conventional approaches to elements present in databases. In this study, we aimed to develop an ML model capable of predicting the adsorption energies of catalyst surfaces containing novel metal elements. To evaluate the model's performance, we proposed and validated an elemental grouping iterative split validation method. Additionally, a range of atomic and structural descriptors related to the catalyst surface were integrated, and the predictive accuracy was improved via feature selection using the Boruta algorithm.