Interpretable Machine Learning for Predicting the Heat Transfer Coefficient in Micro-Channel Flow Boiling

Accurate prediction of heat transfer coefficients in micro-channel flow boiling is essential for advanced thermal management systems. While empirical correlations are grounded in physical mechanisms, they are limited by testing data range, and black-box machine learning models lack physical interpretability, raising concerns about extrapolation. In this study, we applied Explainable Boosting Machine (EBM) — a transparent machine learning algorithm — to predict micro-channel flow boiling heat transfer coefficients using 16,440 experimental data points from 36 publications covering 13 fluids. EBM was benchmarked against Artificial Neural Network (ANN), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), as well as empirical correlations. EBM achieved 14.69% mean absolute percentage error (MAPE), higher than ANN’s 7.12%, but offers superior interpretability. Its additive structure decomposes individual and interactive effects of input variables, elucidating dominant physical mechanisms. The model captured trends consistent with known behaviors, such as heat and mass flux influences and dryout phenomena at high vapor qualities, while revealing parameter interactions beyond empirical or black-box models. A case study based on 243 data points from in-house R410A experiment achieved 7.90% MAPE for EBM, showing robust prediction accuracy with limited dataset. This work demonstrates that interpretable machine learning can complement traditional correlations by offering predictive capability and transparent, physically consistent explanations for micro-channel flow boiling system development and optimization.