Tensile behavior evaluation of two-stage concrete using an innovative model optimization approach

Two-stage concrete (TSC) is a sustainable material produced by incorporating coarse aggregates into formwork and filling the voids with a specially formulated grout mix. The significance of this study is to improve the predictive accuracy of TSC's tensile strength, which is essential for optimizing its use in construction applications. To achieve this objective, novel and reliable predictive models were developed using advanced machine learning algorithms, including random forest (RF) and gene expression programming (GEP). The performance of these models was evaluated using important evaluation metrics, including the coefficient of determination (R 2), mean absolute error (MAE), mean squared error, and root mean square error (RMSE), after they were trained on a comprehensive dataset. The results suggest that the RF model outperforms the GEP model, as evidenced by a higher R 2 value of 0.94 relative to 0.91 for GEP and reduced MAE and RMSE error values. This suggests that the RF model has a superior predictive capability. Additionally, sensitivity analyses and SHapley Additive ExPlanation analysis revealed that the water-to-binder (W/B) ratio was the most influential input parameter, accounting for 51.01% of the predictive outcomes presented in the model. This research emphasizes optimizing TSC design, enhancing material performance, and promoting sustainable, cost-effective construction.