A regularized augmented group contribution method for predicting thermophysical properties of sustainable refrigerants

Abstract Traditional Group Contribution methods (GCMs) cannot reliably distinguish stereoisomers, for example, cis – trans , because molecular spatial information is not explicitly considered. To address this limitation, a molecular eccentricity descriptor, which represents three‐dimensional molecular shape, is introduced to augment the input features of the GCM. By regularizing estimated parameters of the augmented GCM, a Regularized Augmented Group Contribution (RAGC) framework is further proposed to improve parameter stability and predictive performance. The model is evaluated on a dataset of over 350 refrigerants and benchmarked against conventional GCMs and representative machine learning models. Results show that RAGC achieves high predictive accuracy across five thermophysical properties while retaining clear physical interpretability. Incorporating eccentricity substantially enhances structural discrimination, enabling robust cis–trans and constitutional isomer differentiation. Furthermore, Feature‐importance comparisons () further reveal that models with higher predictive accuracy exhibit stronger consistency with RAGC contribution rankings. Overall, RAGC provides a balanced framework combining accuracy, interpretability, and isomer‐distinguishing ability.