Machine learning prediction of thermophysical properties of TiO₂–CuO/water hybrid nanofluid: an SVR–SHAP framework with three-protocol cross-validation

The accurate prediction of thermophysical properties are required to design hybrid nanofluid based thermal systems. In this study, the thermal conductivity and specific heat capacity of TiO₂–CuO (50:50) hybrid nanofluid is studied experimentally in the temperature and volume concentration range of 30–60 °C and 0–0.5 vol%, respectively. From the experimental results, it was found that thermal conductivity was higher with increasing temperature and nanoparticle concentration up to 12.6% at highest concentration (0.5 vol%) and temperature (60 °C) when compared to the base fluid. At 30 °C, water exhibited a 20.41% higher specific heat capacity than nanofluid concentration of 0.5 vol%. The experimental data were then used for building Support Vector Regression (SVR) models for estimation of thermal conductivity and specific heat capacity. Three cross validation protocols were applied to evaluate model robustness and generalization abilities, and SHapley Additive exPlanations (SHAP) analysis was used to explain the influence of input parameters on model outputs. The predictive performance of the developed SVR models was excellent under LOO and LOTO validation, with markedly reduced performance under LOPO (concentration-level) extrapolation, particularly for specific heat capacity. In general, the proposed SVR–SHAP model is an accurate, strong, and interpretable tool to predict the thermophysical properties of TiO₂–CuO/water hybrid nanofluid.