Abstract: Direct coal liquefaction technology is a critical process for converting coal into clean liquid fuels and high-value-added chemicals. As a significant product of the direct coal liquefaction process, naphtha plays a crucial role, and the study of its properties is of great importance for improving product quality and utilization efficiency. In this study, machine learning methods were employed to construct a predictive model for the properties of direct coal liquefaction naphtha. Experimental samples were obtained from a million-ton-per-year direct coal liquefaction plant in Ordos, including hydrogenated naphtha, light naphtha, heavy naphtha, and reformed naphtha. Systematic analysis of the composition and properties of these samples was conducted to establish the dataset required for model training. Three algorithms—Support Vector Regression (SVR), Artificial Neural Network (ANN), and Extreme Gradient Boosting (XGBoost)—were utilized to develop the predictive model, and the SHAP method was applied to interpret the key compositional factors influencing naphtha properties. The results indicate that the presence of C5 and C6 alkanes as well as C7 and C8 cycloalkanes reduces the octane number of naphtha, while an increase in C7 and C8 aromatic content enhances the octane number. The impact of isoalkanes on the octane number exhibits carbon-number dependency. In terms of density, low-carbon-number (C6?) group compositions decrease naphtha density, whereas high-carbon-number (C8+) group compositions increase it. Based on the constructed predictive model, a naphtha property prediction software was developed using the PySide 6 framework combined with Qt Designer tools. This software has been applied in practical industrial settings, demonstrating high consistency between predicted and measured values.