Predicting Band Gap of Carbon and Nitrogen-Based Photocatalysts Using Machine Learning

The increasing need for sustainable energy solutions has driven the development of photocatalytic materials with tailored band gap properties. In this study, a Random Forest Regressor model was developed to predict the band gap of materials incorporating carbon and nitrogen. The model utilized a dataset comprising 3626 materials with features such as density, energy above the hull, magnetic ordering, and structural parameters, with data sourced from Materials Project. The predictive performance of the model was evaluated using metrics including MAE = 54455 eV, RMSE = 54600 eV, and a R = 54.13, indicating strong predictive capability. Feature importance analysis revealed that magnetic ordering and density were the most influential factors, contributing 222 and 1641%, respectively, to band gap predictions. A correlation heatmap further highlighted the relationships among material properties, with density showing a strong negative correlation (-541.) with band gap values. The findings demonstrate the effectiveness of machine learning in accurately predicting band gaps, overcoming the limitations of traditional computational methods, and enabling the rapid identification of promising photocatalysts. This approach significantly accelerates the discovery of materials for applications in water splitting, CO reduction, and environmental remediation, supporting the transition to sustainable energy solutions.

Predicting Band Gap of Carbon and Nitrogen-Based Photocatalysts Using Machine Learning Keywords:

Photocatalysts , Band gap prediction , Machine learning , Carbon and nitrogen-based materials , Random forest regressor