Machine learning enhanced prediction of sensible heat storage potential based on thermogravimetric analysis

Abstract The challenge of efficiently predicting the sensible heat storage potential of natural materials, including the accurate prediction of sensible heat storage capacity in the materials, presents a critical challenge for sustainable thermal energy systems. This study addresses this challenge through a comprehensive machine learning(ML) approach to model the thermal behavior of Dawakin Tofa clay, an abundant and eco-friendly material with potential to replace synthetic thermal storage media. We systematically evaluate twelve predictive models, including four linear approaches (Interactive Linear Regression (ILR), Stepwise Linear Regression (SWLR), Robust Linear Regression (RLR), and Kernel Support Vector Machine (KSVM), four nonlinear methods (G-Matern 5/2 (GM5/2), Trilayered Neural Network (TNN), Boosted Trees (BoT), and Bagged Tree Neural Networks (BTNN), and three ensemble techniques (Simple Average Ensemble (SAE), Weighted Average Ensemble (WAE), and Neural Network Ensemble (NNE). Experimental validation reveals that the NNE demonstrates exceptional predictive performance, achieving near-perfect accuracy with minimal error metrics (MSE = 0.0001696, RMSE = 0.01302 in testing phase). Comparative analysis reveals that SAE offers moderate yet stable generalization (test RMSE = 0.0187), whereas WAE exhibits higher variance (RMSE train-test = 0.0098), suggesting potential overfitting. The superior performance of nonlinear models, particularly NNE, underscores their ability to capture the complex thermal behavior of natural materials. The present work makes three key contributions, namely, a robust ML framework for thermal property prediction, quantitative evidence supporting natural clay as sustainable thermal storage media, and methodological insights into model selection for material science applications. The findings advance renewable energy research by demonstrating how locally available materials can be optimized through computational modeling, with implications for reducing reliance on synthetic alternatives. Future research directions should investigate material modifications, multi-scale validation, and techno-economic analysis to facilitate the practical implementation of the technologies.