Employing ensemble learning techniques for modeling nanofluids' specific heat capacity

Abstract

Accurate investigation of nanofluids' specific heat capacity as an effective thermophysical parameter in applications of heat transfer-based equipment and solar-based systems can play a crucial role in improving their performance and efficiency. The current study utilized advanced machine learning techniques to predict nanofluids' specific heat capacity. This target variable was considered as a function of average nanoparticle size, particle volume fraction, temperature, specific heat of nanoparticles, and specific heat of the base-fluids. In order to predict the specific heat values of different nanofluids, 2084 experimental data related to 10 types of nanoparticles and 6 types of base-fluids were extracted from the literature. In the next step, (i) artificial neural network models such as Multi-Layer Perceptron (by employing different training algorithms) and Radial Basis Function, and (ii) advanced ensemble learning techniques including Categorical Boosting, Decision Tree, Light Gradient Boosting Machine, and Random Forest were implemented on the dataset. Results indicated that the Random Forest model with an Average Absolute Percent Relative Error (AAPRE) of 0.1928% and a Coefficient of Determination (R2) of 0.9995 had the highest accuracy compared to the other ones. Moreover, for the best model (i.e., Random Forest model), sensitivity analysis was performed to determine the relationships between the input and output parameters, and the Leverage statistical technique was applied to find the valid, outlier, and out-of-range data-points.