Machine learning predicts strength of nanoparticle-modified basalt fiber composites

Machine learning models can now predict the tensile and flexural strength of nanoparticle-modified basalt fiber reinforced epoxy composites, according to a peer-reviewed study published in Discover Materials on 27 October 2026. The research, authored by Md. Saphin Hasan and Md. Ashraful Islam, built eight supervised regression models trained on two experimental datasets compiled from previously published studies.

Basalt fiber reinforced epoxy composites are widely studied structural materials, and adding nanoparticles to the epoxy matrix is a known strategy for improving overall mechanical performance. According to the study, nanoparticles enhance matrix stiffness, the bond between fibers and matrix, crack resistance, and stress transfer efficiency. However, the final strength of such a composite depends on many interacting variables, including the type of nanoparticle, particle ratio, particle size, curing condition, fiber layer count, fiber orientation, fiber density, the test standard used, and the manufacturing process. That complexity makes traditional trial-and-error experimentation slow and costly, which is the gap the authors set out to address with data-driven prediction.

The researchers trained and evaluated eight algorithms: Linear Regression, LightGBM, AdaBoost, Gradient Boosting, K-Nearest Neighbors, CatBoost, Extreme Gradient Boosting, and Random Forest.

Performance differed by target property. For tensile strength, CatBoost was the best predictive model, achieving a testing R² of 0.8530 and a testing mean absolute error of 58.39 MPa. For flexural strength, Random Forest delivered the strongest results, with a mean absolute error of 44.94 MPa and an R² of 0.9512.

A key contribution of the study is its emphasis on interpretability rather than prediction alone. The authors incorporated partial dependency plots, which show how predicted strength changes as an input variable varies, and SHAP (SHapley Additive exPlanations) analysis, which quantifies each feature’s contribution to individual predictions. Together, these tools reveal which parameters matter most and whether their effects are linear, nonlinear, or dependent on category.

The SHAP analysis indicated that fiber orientation, curing temperature, and test standard had the largest influence on tensile strength. Flexural strength, by contrast, was strongly influenced by a broader set of factors: fiber orientation, particle type, test standard, particle size, and fiber layer. The partial dependency plots showed nonlinear effects and category-dependent behavior for important numerical and categorical variables, meaning that the effect of a given parameter can change in magnitude and direction depending on its value or class.

To probe robustness, the authors applied a leave-one-study-out (LOSO) evaluation, and the results showed the models were robust.

The authors describe the proposed framework as a tool that can predict the mechanical strength behavior of nanoparticle-modified basalt/epoxy composites, helping to reduce experimental trial and error and supporting the design of optimized composite materials. Because the models were trained on published experimental data rather than new laboratory measurements, their accuracy reflects the scope and quality of the compiled datasets, and the study presents the framework as a design aid rather than a replacement for experimental verification.

For basalt/epoxy composites modified with nanoparticles, the study's SHAP analysis points to fiber orientation, curing temperature, and test standard as the largest influences on tensile strength, and to fiber orientation, particle type, test standard, particle size, and fiber layer as strong influences on flexural strength.

Subject of Research: Interpretable machine learning prediction of tensile and flexural strength in nanoparticle-modified basalt fiber reinforced epoxy composites

Article Title: Interpretable machine learning-based prediction of tensile and flexural strength of nanoparticle modified basalt fiber reinforced epoxy composites

Article References: Hasan, M. S., & Islam, M. A. (2026). Interpretable machine learning-based prediction of tensile and flexural strength of nanoparticle modified basalt fiber reinforced epoxy composites. Discover Materials. https://doi.org/10.1007/s43939-026-01030-6

Image Credits: AI Generated

DOI: 10.1007/s43939-026-01030-6

Keywords: machine learning, basalt fiber composites, epoxy, nanoparticles, tensile strength, flexural strength, SHAP, CatBoost, Random Forest, interpretability, materials design, Interpretable