Standard
enthalpy of formation is an important fuel thermochemical
property for energy balance calculation and equilibrium dynamics determination
in combustion reactions. In this study, an approach combining the
quantitative structure–property relation and the active subspace
method was applied to predict the standard enthalpies of formation
of different hydrocarbon classes, including alkanes, cycloalkanes,
alkenes, alkynes, and aromatics. The standard enthalpies of formation
in a database of 1020 hydrocarbons were predicted by a one-dimensional
active subspace model, with 902 molecule topological indices being
used as input descriptors. The correlation coefficient between the
measured and predicted values was 0.99, and the average absolute error
was 6.74 kJ/mol. The sensitivity of the prediction output to each
descriptor was further evaluated, which enables us to build simplified models by
only considering the most representative descriptors. Also, to assess
the influences of model simplification on the prediction performance,
the relation between the descriptor number and the prediction accuracy
was explored. It was observed that simplified but well-predictive
models can be established when the top 10 influential descriptors
were selected for model building, and the correlation coefficient
between the predicted and measured standard enthalpies of formation
was as high as 0.96.