Machine Learning-Based Predictions of Nanofluid Thermal Properties

Abstract

In this study, machine learning-based predictions of thermal conductivity, dynamic viscosity, and specific heat of nanofluids are explored. Various types of nanofluids and parametric conditions are considered to broaden and evaluate the effectiveness of popular machine learning models, including multilayer perceptron, random forest, light gradient boosting machine, extreme gradient boosting, and stacking algorithms. The performance of these prediction models is assessed using the mean squared error and the coefficient of determination. The influence of each input variable on model development was examined to identify key features. Information gain is introduced and calculated for determining the importance of parameters in prediction. External validation is performed with an additional unseen dataset to further assess the applicability of the selected models across different experimental data points. It was found that the stacking technique is the most accurate machine learning algorithm among those investigated. The LightGBM is the top choice when considering both computational accuracy and efficiency. The results demonstrate that machine learning methods can provide excellent predictions of the thermophysical properties of complex nanofluids.