Machine learning analysis of catalytic CO2 methanation

Abstract

In this work, a detailed dataset containing 4051 data points gathered from 527 distinct experiments in 100 published articles for catalytic CO2 methanation was analyzed using machine learning methods. A pre-analysis of the database was performed using simple descriptive statistics while a random forest (RF) model was developed to predict CO2 conversion as the function of 23 descriptors including catalyst properties, preparation methods, and reaction conditions. Boruta analysis was also performed to identify the significant variables. The random forest model was found to be quite successful in predicting CO2 conversion with the training and testing root mean square error (RMSE) of 6.4 and 12.7 respectively; R2 was 0.97 for training while it was 0.85 for testing. The success of the model was also verified by computing CO2 conversion profiles for individual experiments in test data and comparing them with those reported in the related papers.