Predicting global warming potential: A self-training and ensemble machine learning approach

Abstract

Amid growing concerns about global warming, a critical challenge for the semiconductor industry is the development of materials with low global warming potential (GWP). However, determining GWP is complicated for newly developed compounds, making the assessment of their environmental impact difficult. Instead of relying on experimental determination, Chung’s group pioneered the use of machine learning to predict GWP values [G. Zhao, H. Kim, C. Yang, and Y. G. Chung, J. Phys. Chem. A 128, 2399 (2024)]. In this study, we propose a machine learning approach named ST-GWP modeling, which combines self-training with an ensemble learning technique to improve prediction accuracy. The model achieved an R2 of 0.97 on the training set and 0.92 on the test set in the regression task. Specifically, in the classification task, our model achieved an accuracy of 0.99 on the training set and 0.96 on the test set, demonstrating robust generalization capabilities. Furthermore, the analysis suggests that GWP is influenced by various molecular properties, including the number of halogen atoms and the distribution of electrons within the molecule, such as polarity, polarizability, and partial charge.