Machine learning-based prediction of the adsorption energy of CO on boron-doped graphene

Abstract

In recent years, doped graphene has been often used as a gas-sensitive material, and its gas adsorption energy has received extensive attention. However, the numerous types of doped graphene make the simulation calculation study of their adsorption energy highly costly and time consuming. In this study, in order to accurately and quickly examine the CO adsorption capacity of different boron-doped graphene structures, 1864 different sets of CO adsorption energy on boron-doped graphene were obtained by simulation based on density functional theory (DFT), and an overall framework based on machine learning was then proposed. After that, three different machine learning methods were evaluated, and the random forest method was found to perform best, with the average root mean square error achieving around 0.051. This study demonstrates the power of machine learning models in uncovering complex and hidden structure–property relations in boron-doped graphene and provides the possibility of using a data-driven method for the rational design of gas-sensitive materials.