Machine learning predictions of onset and oxidation potentials for methanol and ethanol electrooxidation: Comprehensive analysis and experimental validation

The onset and oxidation potentials of electrochemical reactions are pivotal in assessing catalytic energy efficiency, spanning applications across various domains, including sustainable energy generation. However, predicting these potentials presents a complex and uncharted challenge. In this study, we present a pioneering approach to developing predictive models for the onset and oxidation potentials within electrochemical reactions linked to the oxidation of methanol and ethanol. We have devised a comprehensive pipeline from Data Collection, Information Extraction, and Preprocessing and assessed the performance of different regression models: Linear, Random Forest, and XGBoost. For the oxidation potential prediction, an RMSE of 0.169 and an R${}^2$ value of 0.814 were achieved. Similarly, for the onset potential prediction, the model yielded an RMSE of 0.185 and an R${}^2$ value of 0.839. The models were further evaluated using feature importance and SHAP values, enhancing our understanding of their predictive mechanisms and providing more comprehension of the features. Additionally, we conducted experimental validations by comparing the predicted outcomes to actual results obtained from methanol and ethanol oxidation experiments carried out in a chemical laboratory. This validation process included the utilization of platinum, gold, nickel foam, steel and RuO${}_2$/FTO electrodes. Encouragingly, the experimental validation yielded promising findings, exhibiting an RMSE of 0.0967 for the onset potential and an RMSE of 0.0234 for the oxidation potential.