XGBoost-based prediction of heat and mass transfer in phase change systems: Insights from ISPH simulation and parameter sensitivity analysis

This study presents a hybrid numerical–machine learning framework for predicting heat and mass transfer behavior in porous cavities filled with Nano-Encapsulated Phase Change Materials (NEPCMs), which are increasingly used in thermal management, energy storage, and microfluidic systems. The purpose is to develop a fast and accurate surrogate model using eXtreme Gradient Boosting (XGBoost) trained on high-fidelity data from Incompressible Smoothed Particle Hydrodynamics (ISPH) simulations. The physical domain considered is a 2D porous cavity with embedded NEPCM and internal cooling zones. The influence of cooling zone length ( L x = 0.1–0.4), cooling zone height ( Lᵧ = 0.1–0.4), and Darcy number (Da = 10 -5 –10 -2 ) on the average Nusselt ( ) and Sherwood ( ) numbers is examined. Results show that extending L x enhances thermal uniformity by up to 17.2%, and increasing Da significantly improves convective transport, with rising by 138.5%. The XGBoost model achieves a Mean Squared Error (MSE) of 1.2×10 -5 and R 2 > 0.98 in all prediction scenarios. The study recommends using data-driven modeling to accelerate parametric optimization in NEPCM-based systems. This framework reduces simulation time while maintaining high accuracy, outperforming traditional CFD-only approaches and providing a flexible platform for future system-scale design and control.