<p>This study presented a predictive optimization framework for evaluating the thermophysical properties of multi-walled carbon nanotube NFs dispersed in a 50:50 water–ethylene glycol base fluid. The main objective was to simultaneously predict TC and dynamic µ_nf, addressing a key limitation of previous studies that focused primarily on TC alone. A feedforward artificial neural network with two hidden layers was developed and validated using experimental data. The dataset covered nanoparticle volume concentrations between 0.025% and 0.1% and temperatures ranging from 25&#xa0;°C to 80&#xa0;°C. The proposed model demonstrated strong predictive capability across all evaluation metrics. Under 10-fold cross-validation, the root mean square error for TC varied from 1.31 × 10⁻⁴ to 3.71 × 10⁻⁴ W/m·°C, while the corresponding values for µ_nf range from 0.010 to 0.031 mPa·s. Low mean-squared error values across the training, validation, and test datasets confirmed the robustness of the learning process. Optimal performance was achieved at epoch 5 for TC and at epoch 8 for µ_nf. In all cases, the coefficient of determination exceeded 0.99, indicating excellent agreement between predictions and experimental measurements. Relative errors remained limited to 0.32–1.57% for TC and 0.12–0.25% for µ_nf, while absolute errors were also tightly bounded. A complementary sensitivity analysis further supported the model stability. A 10% variation in Temperature led to maximum deviations of 2.636% in TC and 0.623% in µ_nf, whereas the same variation in nanoparticle concentration produced larger deviations of 5.744% and 0.893%, respectively. Despite this difference, mean deviations remained modest for both properties, confirming the robustness of the proposed framework under input perturbations.</p>