QSPR-based prediction of critical and molecular parameters for radical and intermediate species toward virial equation of state modeling

The lack of critical properties, dipole moments, and acentric factors for radicals and intermediates limits the application of the virial equation of state (EoS) in combustion modeling. In this work, quantitative structure–property relationship (QSPR) models were developed to predict the critical temperature (T_c), critical pressure (p_c), dipole moment (μ), and acentric factor (ω) of radical and intermediate species. Datasets for the four target properties were collected from 83 compounds with available property data, and a correlation-controlled descriptor screening strategy was used to retain 6 to 8 molecular descriptors for each property. Support vector regression (SVR), back-propagation neural network (BPNN), and least-squares boosting (LSBoost) models were constructed and compared. The BPNN models showed the best overall predictive performance, with R_ext² values of 0.939, 0.956, 0.949, and 0.944 for p_c, T_c, μ, and ω, respectively. External validation and applicability domain analysis further confirmed the stability and robustness of the developed models. The proposed QSPR framework provides a practical route for generating provisional estimates of missing parameters for radicals and intermediates and may support future applications of the virial EoS in combustion simulations. Comparison with conventional group-contribution methods indicates that the descriptor-based QSPR model offers certain advantages in species coverage, although its impact on downstream combustion predictions requires further validation in future work.