AI-Based Modeling and Predictive Analysis for Ternary Hybrid Nanofluids (THNFs) for Advanced Thermal Applications

This study presents a comparative machine learning framework for predicting the thermal conductivity (k), electrical conductivity (EC), viscosity (μ), and pH of magnetic (Fe₃O₄/Al₂O₃/TiO₂) and non-magnetic (Fe₂O₃/Al₂O₃/TiO₂) ternary hybrid nanofluids as a function of temperature and nanoparticle concentration. Two regression algorithms were trained and evaluated on the same feature set: Gaussian Process Regression (GPR), a non-parametric Bayesian approach that models the underlying function as a distribution over possible fits, and Random Forest (RF) regression, an ensemble tree-based baseline. Model performance was assessed using the coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE) for each of the four target properties. Gaussian Process Regression achieved higher R² and lower error than Random Forest for three of the four properties — thermal conductivity (R² = 0.797 vs. 0.770), viscosity (R² = 0.949 vs. 0.803), and pH (R² = 0.945 vs. 0.821) — while Random Forest achieved a marginally higher R² for electrical conductivity (0.803 vs. 0.787). These results indicate that Gaussian Process Regression is generally the stronger predictive model for this dataset, particularly for viscosity and pH, where it reduced RMSE by roughly half relative to Random Forest, while the two algorithms are effectively comparable for electrical conductivity. The findings support the use of Gaussian Process Regression as the primary surrogate model for thermophysical property prediction in ternary hybrid nanofluids, with Random Forest retained as a useful, more interpretable baseline for comparison.