Comparative Analysis and Optimization of Machine Learning Algorithms for Prediction of Adsorption Energy of Methane Related Species on Cu-based Alloys

Abstract

Demand of ethylene is surging at unprecedented rate and therefore conversion of methane into these important value-added chemicals using one-step processes such oxidative coupling of methane is of paramount importance. It would be beneficial and significant to design a catalyst which can suppress undesired over reactions which lead to the production of COx gases resulting in issues of yield and selectivity. To this end, in this study, Machine learning (ML) models integrated with genetic algorithm (GA) have been developed to predict, evaluate, and analyze adsorption energies of methane related species on Cu-based Alloys. Comparative study of different ML algorithms integrated with Genetic Algorithm (GA) were performed to improve the ML model’s architecture and parameters selection. The results proposed that Categorical boosting (Catboost) model outperformed all other models and effectively predicts adsorption energies compared to other models (RMSE = 0.0977, CC = 96.5 %). Permutation importance score was utilized to asses prediction performance and accuracy which revealed that group number and surface energy contributed the highest to the model. The partial dependence plots (PDPs) analysis shows the potential effects of each influencing parameter impact on the prediction of the respective adsorption energies and as well as shows that how these factors will interact during oxidative coupling of methane (OCM). Finally, in order to analyze the distribution of the data points of the features and how they affect the model’s predictions, bee swarm plot was employed.