predicting water-based nanofluid viscosity using machine learning: performance evaluation of ls-svm, gmdh, and gp models

Viscosity is one of the most important properties of nanofluids, which has a direct influence on fluid flow as well as convective heat transfer. despite extensive experimental studies and theoretical models, accurate prediction of nanofluid viscosity remains challenging due to the complex interplay of nanoparticle properties, base fluid characteristics, and temperature effects. herein, three machine learning approaches, including least squares support vector machine (ls-svm), group method of data handling (gmdh), and genetic programming (gp), were designed and compared for predicting the effective viscosity of water-based nanofluids containing al2o3, cuo, tio2, and sio2 nanoparticles (nps). for this purpose, a collection of 795 experimental data points has been extracted from literature sources. the effective input variables, including temperature, the viscosity of water (as the base fluid), volume fraction, and diameter of the nps, were used as input variables of the models. to find the best model for predicting nanofluid viscosity, the performance of the developed models was evaluated via statistical and graphical methods. in addition, the prediction ability of the machine learning methods was compared with some well-known theoretical models. the obtained results showed that the ls-svm model can predict the experimental data with an average absolute relative deviation of 1.571% and a coefficient of determination up to 0.9952. accordingly, the ls-svm model showed the best performance and the most reliable and accurate prediction of water-based nanofluid viscosity. the shapley additive explanation (shap) method was also applied to predict the influence of each data point and each input feature on the model output, revealing the significance of the volume fraction of nps and viscosity of water with absolute values of 1.65 and 0.25, respectively.