Predicting tensile strength of steel fiber-reinforced concrete based on a novel differential evolution-optimized extreme gradient boosting machine

Splitting tensile strength (f_spt) is a crucial parameter in designing concrete mixes. The addition of steel fibers helps improve the mechanical properties of concrete, including its f_spt. Due to the complexity and time-consuming processes required in conducting tensile tests, this study proposes a novel integration of differential evolution (DE) and an extreme gradient boosting machine (XGBoost) for estimating the f_spt of concrete mixes based on their constituents, age, and compressive strength. XGBoost is used to learn the nonlinear and multivariate mapping function between the f_spt and its influencing factors. In addition, DE, as a metaheuritic algorithm, is employed to automatically optimize the XGBoost performance. A dataset, including eight predictor variables and 173 records, is used to train and verify the hybrid DE-XGBoost approach. Experimental results, supported by statistical tests, show that the newly proposed method can achieve outstanding predictive accuracy with a mean absolute percentage error (MAPE) of 9.5% and a coefficient of determination (R²) of 0.9. Sensitivity analysis shows that the contents of aggregates and water critically affect the f_spt. Meanwhile, the volume fraction of steel fiber, aspect ratio of steel fibers, binder quantity, and concrete age moderately influence tensile strength performance. Moreover, an asymmetric squared error loss function is used during the training phase of XGBoost to reduce the percentage of overestimated f_spt values from 52 to 35% with a minor loss of predictive accuracy.