APPLICABILITY OF MACHINE LEARNING TECHNIQUES IN PREDICTING SPECIFIC HEAT CAPACITY OF COMPLEX NANOFLUIDS

The complexity of the interaction between base fluids and nano-sized particles makes the prediction of nanofluid thermophysical properties difficult. However, machine learning techniques can be utilized as an alternative approach due to their ability to identify complex nonlinear patterns in data and make accurate forecasts. This paper presents intuitive predictions of specific heat of various types of nanofluids using machine learning models based on experimental data obtained from 47 different studies, comprising 5009 data points. Three machine learning algorithms, namely, artificial neural network (ANN), support vector regression (SVR), and extreme gradient boosting (XGBoost), were tested to develop a universal predictor for nanofluid specific heat. To enhance the performance of the machine learning models, the best set of input variables was selected, and hyperparameter optimization was conducted to maximize the prediction accuracy. The accuracy of three selected machine learning models [i.e., MLP (a type of ANN), SVR, and XGBoost] and their unseen data prediction capability were compared with existing complicated empirical models, and the results showed that the machine learning-based predictions were more accurate. The machine learning models demonstrated excellent agreement with experimental nanofluid specific heat data. Particularly, the extreme gradient boosting method (i.e., XGBoost) showed the best nanofluid specific heat forecast results with minimal prediction error and presented broad range of applicability.