An overview of artificial intelligence techniques for predicting thermophysical characteristics of nanofluids in heat transfer applications

Nanofluids, which are composed of nanoscale particles of solid dispersed into conventional base fluids, have been given much attention as advanced heat transfer media for various applications such as heat exchangers, solar thermal collectors, electronic cooling systems, heat exchangers and heat management in heat ventilation and air conditioning technologies, lubrication, and biomedical thermal management. The performance of nanofluids is determined by important thermophysical property such as thermal conductivity, viscosity, specific heat capacity and density which prediction is still difficult due to the complex nonlinear interactions between temperature, nanoparticle concentration, particle morphology and base fluid properties. The typical empirical correlations also exhibit the poor applicability and generalization in various nanofluid systems. Artificial intelligence (AI) and machine learning (ML) techniques are popular techniques that can model complex relationships such as this directly from the data. This study is aimed at a systematic literature review carried out following the protocol of PRISMA based to evaluate the AI-driven prediction methods for nanofluid thermophysical properties within the time period spanning from 2015 to 2025. One hundred peer-reviewed journal articles were analyzed and categorized according to nanofluid type, thermophysical properties under investigation, data sources and AI methodologies. Artificial neural network are the most commonly used models, and ensemble learning methods like random forest and XGBoost are getting more robust and accurate in terms of prediction. Key research gaps include insufficient amount of experimental validation, lack of experimental research involving hybrid and ternary nanofluids, lack of adoption of explainable and physics-informed AI frameworks. Future research directions toward unified, interpretable and application-ready predictive models are discussed.