Machine learning-based prediction of compressive strength of wastewater-exposed concrete using explainable and optimized models

Reinforced concrete structures are extensively used in wastewater collection systems; however, their structural performance can progressively deteriorate under aggressive sewer environments associated with gas generation and chemical attack. Conventional empirical approaches for estimating residual compressive strength may be time-consuming and may not adequately capture complex nonlinear relationships among influencing parameters. The compressive strength of concrete subjected to wastewater was therefore predicted using optimized and interpretable machine learning algorithms. Four nonlinear models Extreme Gradient Boosting (XGBoost), Gene Expression Programming (GEP), Light Gradient Boosting Machine (LightGBM), and Support Vector Regression (SVR)were integrated with Particle Swarm Optimization (PSO) for hyperparameter optimization, while Multiple Linear Regression (MLR) was employed as a conventional benchmark. A dataset comprising 138 experimental records with five (5) inputs was divided into training and testing subsets (80/20), and all model performance was assessed using R 2 , RMSE, NSE, a-20 index, 10-fold cross-validation, and uncertainty analysis. Among the developed models, PSO-XGboost exhibited the best testing performance (R 2 = 0.960), RMSE = 1.25 MPa, NSE = 0.94, and a-20 = 0.97, followed by PSO-GEP (R 2 = 0.950), RMSE = 3.21 MPa. PSO-LightGBM, MLR, and PSO-SVR achieved testing (R 2 ) values of 0.891, 0.892, and 0.880, with RMSE values of 4.92, 7.34, and 5.41 MPa, respectively. Regarding the novelty, a prototype Graphical User Interface (GUI) framework was developed to facilitate practical compressive-strength estimation. Additionally, a GEP-based predictive equation has been developed for assessing the compressive strength.