Predictive, correlative and machine learning models for estimation of viscosity of liquid mixtures

Among the various thermodynamic and transport properties, viscosity is a crucial thermophysical property that plays a vital role in heat and mass transfer, design and calculations, research and development, and numerous industrial applications. Predicting the viscosity behaviour of this highly significant transport property in liquids and liquid mixtures has been a challenging task due to its sensitivity to temperature and concentration variations. In the current investigation, 42 binary liquid mixtures have been investigated and 9 predictive and 4 correlative approaches, along with the machine learning algorithm - Random Forest, have been utilized to predict the viscosity at varying temperatures and concentrations. Percentage Average Absolute Deviation (%AAD) has been used to carry out a comparative analysis to assess the predictive efficacy of these different methodologies. The findings indicate that the Dey-Biswas model outperforms all the predictive models with a Grand %AAD value of 3.41. The Grunberg-Nissan and Wijk give the best overall performance in correlative models having a Grand % AAD of 1.17 with Random Forest also demonstrating consistent predictive efficacy with better Grand %AAD values.