Machine learning algorithms to predict the tensile strength of novel composite materials

Natural fiber-reinforced polymer (NFRP) composites are increasingly used in structural, automotive, and aerospace applications due to their lightweight, sustainable, and cost-effective properties. A major challenge in their development is the lack of experimental tensile strength data for new formulations. This study develops machine learning (ML) models to predict NFRP tensile strength using publicly available datasets containing parameters such as epoxy group content, density, elastic modulus, curing agent amount, resin consumption, surface density, and matrix–filler ratio. Five regression algorithms such as polynomial regression, bagging regression, random forest, XGBoost, and gradient boosting were trained and evaluated using five-fold cross-validation and standard error metrics. Unlike prior works, this approach systematically removes weakly correlated features to enhance both accuracy and interpretability. Random forest delivered the highest performance (R² = 0.92, mean absolute error (MAE) = 1.64), outperforming conventional and ensemble models. The results establish a reproducible, data-driven framework capable of reducing experimental workloads and accelerating sustainable composite design. Future studies will extend this methodology to multi-property prediction, integrating microstructural descriptors for improved real-world applicability.