<p>Accurate prediction of hybrid nanofluid density is critical for the efficient design of thermal energy systems. In this study, four tree-based machine learning algorithms (Random Forest (RF), Decision Tree (DT), Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM)) were systematically developed and compared to predict hybrid nanofluid density. A dataset consisting of 436 experimental data points compiled from the literature was used, including nanoparticle type, base fluid type, temperature, volumetric concentration, and component densities as input parameters. An 80/20-layer training/test split was applied and fivefold cross-validation was performed to evaluate generalization performance. All models achieved test set <i>R</i>² values exceeding 0.95; XGBoost gave the best performance (<i>R</i>² = 0.9906, MAE = 1.30&#xa0;kg&#xa0;m⁻³, RMSE = 2.443&#xa0;kg&#xa0;m⁻³, MAPE = 0.13%). Cross-validated <i>R</i>² means ranged from 0.9577 (DT) to 0.9885 (XGBoost); The maximum difference between cross-validated and test set <i>R</i>² did not exceed 0.009 in all models, confirming stable generalization. XGBoost’s superior accuracy is attributed to its built-in <i>L</i>1/<i>L</i>2 regularization and iterative residual error correction mechanism. Feature importance analysis identified base fluid density and nanoparticle volumetric concentrations as the dominant determinants of hybrid nanofluid density. The proposed models offer a fast and low-cost alternative to experimental density measurement; however, their applicability is limited to the parameter ranges of the training dataset.</p>