SISSO‐Based Interpretable Machine Learning for Predicting Jet Fuel Properties From GC × GC Data: A New Approach to Structure–Property Relationship Modeling

Jet fuel is a complex mixture containing hundreds to thousands of compounds, which poses significant challenges for accurate property prediction and rational formulation. In this study, a comprehensive dataset of 131 jet fuel samples was characterized using comprehensive 2D gas chromatography (GC × GC‐FID/MS), yielding a 131 × 91 feature matrix and six properties of the fuels. A sparse identification of selected operators (SISSO) framework is applied to establish interpretable symbolic models, generating analytical expressions that directly link molecular families (n‐alkanes, iso‐alkanes, cycloalkanes, aromatics, etc.) to macroscopic fuel properties. Compared with conventional black‐box machine learning models, SISSO not only achieves competitive predictive accuracy (R 2 up to 0.94 with minimal RMSE) but also provides physical interpretability. The mathematical models demonstrate that branched alkanes and aromatics significantly reduce viscosity, freezing point, and flash point, while long‐chain n‐alkanes increase density and freezing point. This work highlights the innovative role of interpretable machine learning in fuel design and offers a transparent framework for structure–property relationship exploration in complex hydrocarbon systems.