Machine learning models to predict sewer concrete strength exposed to sulfide environments: unveiling the superiority of Bayesian-optimized prediction models

The climate catastrophe is the result of unrelenting developments and overuse of concrete resources. One development is the reinforced concrete sewerage network, which needs constant maintenance due to its strength deteriorates over time. Therefore, this study approaches the development of optimized, unoptimized, and standalone models such as extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), random forest (RF), artificial neural network (ANN), and support vector machine (SVM) to predict the concrete compressive strength exposed to aggressive conditions. Therefore, this study and concrete materials must use ensembles and non-linear models. Accordingly, 138 specimens with five significant inputs were gathered for training (80%) and testing (20%) assessment. K-fold cross-validation and fitness functions (R², RMSE, MAE, PI, VAF, LMI, and a-20 index) were used to evaluate the models' dependability. Among all techniques, the ensemble optimized method (BO-XGboost) provided a remarkable correlation (R² = 0.99) in comparison to the rival methods such as BO-Ad boost (R² = 0.94), BO-RF (R² = 0.93), XGBoost (R² = 0.93), Adboost (R² = 0.91), RF (R² = 0.90), ANN (R² = 0.82), SVR (R² = 0.74). Similarly, the BO-XGBoost exhibited the slightest error of RMSE = 1.87 MPa among all techniques. Furthermore, the k-fold cross-validation confirmed the model's (BO-XGB) authenticity and performance. In addition to that, the XGBoost Shapley analysis showed that w/c is the dominant component in the concrete strength prediction. This study's findings verified that w/c has a greater influence on concrete compressive strength compared to other elements. In terms of innovation, the application's graphical user interface is established to make the model simpler for inputs and necessary results and open up a useful and practical engineering route.