Polyethylene glycol-nano composite thermal conductivity: Introduction of data-driven models

The thermal conductivity of polyethylene glycol (PEG)-nanocomposites is crucial for their use in thermal management systems, underscoring the need for accurate predictive models. This study develops a Gradient Boosting Machine (GBM) model optimized with four algorithms: Batch Bayesian Optimization (BBO), Evolution Strategies (ES), Bayesian Probability Improvement (BPI), and Gaussian Processes Optimization (GPO). A dataset of 229 experimental samples (90% training, 10% testing) was employed, using temperature, PEG molecular weight, nano concentration, and nano type as input variables. K-fold cross-validation minimized overfitting. Model performance was evaluated with R-squared (R²), mean squared error (MSE), average absolute relative error (AARE%), and runtime. Correlation analysis showed nano concentration had the strongest impact on conductivity (0.75), followed by PEG molecular weight (0.56), while temperature (0.33) and nano type (0.24) were weaker predictors. Among the algorithms, GBM-ES achieved the best accuracy (R² = 0.9966 training, 0.9158 testing), outperforming the other methods and traditional linear regression, which showed far lower precision. GPO was the most computationally efficient (153.6 s), while BBO was slowest (274 s). Sensitivity analysis confirmed the dominant role of nano concentration and molecular weight. Overall, the optimized GBM models provide reliable, data-driven tools for predicting PEG-nanocomposite conductivity, reducing dependence on costly experimental work.