Prediction of critical properties and boiling point of fluorine/chlorine-containing refrigerants

• Tailored method predicting properties of F/Cl-containing refrigerant is proposed. • The present models outperform the conventional methods in calculation. • An interpretable method was adopted to interpret the proposed models. In this work, molecular groups were used as the descriptor of molecular structures, combining with multi-layer perceptron algorithm to establish the prediction models of boiling point, critical temperature and critical pressure of fluorine/chlorine-containing refrigerants. The absolute average deviation values of the three prediction models between the calculated and experimental data in total datasets are 2.93%, 3.13% and 4.46%, respectively. Furthermore, in order to understand how the group features influence the prediction of properties, an interpretable method, Shapely Additive exPlanation (SHAP) was employed to interpret the proposed models, which computed the contributions of group features, and verified that the present models effectively building the relationship between the molecular groups and thermophyical properties.