XGBoost algorithm for predicting heat transfer coefficient of saturated flow boiling in mini/micro-channels

Abstract

Accurate prediction of the heat transfer coefficient in saturated flow boiling within mini/micro-channels is the most critical factor in designing thermal systems for high-heat-flux devices. This study proposes a machine learning technique to predict the heat transfer coefficient of saturated flow boiling using the XGBoost (eXtreme Gradient Boosting) algorithm. The database used in this study consists of 11,470 pre-dryout data points, obtained by removing 1878 post-dryout data points from a total of 13,348 data points collected from 41 sources, employing an XGBoost incipience dryout predicting model. The dataset includes 23 working fluids, hydraulic diameters ranging from 0.19 mm to 6.50 mm, mass flow rates from 19.45 kg/m²s to 1608 kg/m²s, and saturation temperatures from -40 °C to 201.37 °C. The permutation feature importance (PFI) and SHapley Additive exPlanations (SHAP) values were used for feature selection, while Optuna was used for hyperparameter tuning. A total of seven training features— Pr<inf>f</inf>, x<inf>di</inf>, P<inf>red</inf>, Fr<inf>fo</inf>, Bo, Pr<inf>g</inf> , and Fr<inf>tp</inf> —were selected and used to develop the model. The model achieved a mean absolute error (MAE) of 7.18 %, demonstrating superior predictive performance compared to existing empirical correlations and other machine learning algorithms. This result confirms that XGBoost is an effective and reliable algorithm for predicting the heat transfer coefficient of saturated flow boiling in mini/micro-channels. © 2025 Elsevier Ltd.