ML-enhanced prediction and thermomechanical evaluation of eco-friendly concrete with waste epoxy composites after elevated temperature exposure

This study evaluated the impact of incorporating waste epoxy composites into concrete and developed advanced predictive models to assess thermomechanical performance. Elevated temperatures of 200 °C, 400 °C and 600 °C were applied before testing. Workability and the compressive strength (CS), splitting tensile strength (STS) and flexural strength (FS) tests were evaluated. The concrete strength, measured by CS, STS and FS, decreases as the percentage of waste epoxy composites increases from 0% to 50%. In addition to empirical modelling, four machine learning algorithms including Extreme Learning Machine (ELM), Gaussian Process Regression (GPR), Random Forest and Extreme Gradient Boosting were employed to predict CS based on the epoxy ratio and elevated temperature. The ELM model achieved the highest accuracy with R2 = 0.961 and Root mean square error (RMSE) is 0.364 MPa, followed closely by GPR (R2 = 0.965, RMSE = 0.349 MPa). Shapley Additive Explanation (SHAP) analysis confirmed temperature as the dominant degradation factor, while bootstrap validation demonstrated robust prediction consistency. k-fold and Leave-One-Temperature-Out validations confirmed strong generalisation across thermal regimes. Finally, to estimate the capacities, three equations were developed for concrete with waste epoxy composites exposed to the elevated temperatures.