• This study presents a data-driven ML framework for Nusselt Index prediction. • The Random Committee model achieves superior accuracy with an R of 0.986. • Sensitivity analysis identifies Reynolds number as the most critical parameter. • The method outperforms traditional correlations using key geometric parameters. • This provides a reliable tool for plate heat exchanger design and optimization. Accurate prediction of the Nusselt Index is critical for the thermal design and optimization of plate heat exchangers (PHEs). While traditional empirical correlations exist, they are often constrained by limited geometric parameters and narrow Reynolds number ranges, compromising their accuracy and generalizability. This study introduces a robust, data-driven framework utilizing ensemble machine learning (ML) to overcome these limitations. The models were trained and tested on a comprehensive dataset of 1,119 experimental observations, with input features encompassing key geometric parameters—Chevron angle, aspect ratio, and surface magnification factor—alongside the Reynolds number. A comparative analysis of three tree-based ML algorithms was conducted: Random Committee (RC), Random Forest (RF), and Random Tree (RT). The results demonstrate that the RC ensemble model significantly outperforms the others, achieving exceptional prediction accuracy on the test data with a correlation coefficient (R) of 0.986, a mean absolute error (MAE) of 2.76, and a root mean square error (RMSE) of 7.75. Cross-validation and sensitivity analysis confirmed the RC model’s superior stability, generalizability, and robustness, identifying the Reynolds number as the most influential parameter. This research establishes that ensemble ML models, particularly the Random Committee algorithm, provide a highly accurate and reliable data-driven alternative to conventional correlations for forecasting the Nusselt Index across a wide spectrum of PHE designs and operating conditions.
Alavi et al. (Wed,) studied this question.