Accurately predicting the mechanical behavior of deteriorated reinforced concrete components using natural intelligence-integrated Machine learners

Corrosion in reinforced concrete components promotes the degradation of structural durability during the service life of buildings. In this paper, the performance of several advanced prediction and ensemble machine learning models is compared in terms of their respective abilities to predict the strength of deteriorated reinforced concrete components. The models compared include three single models, namely least square support vector regression (LSSVR), radial basic function neural network (RBFNN), and multivariate adaptive regression splines (MARS), and two ensemble models, namely logit boosting (LogitBoost) and the extreme gradient boosting technique (XGBoost). All machine learning models were trained using an equilibrium optimizer (EO). A dataset consisting of 140 data samples collected from residential buildings in southern Vietnam was used to establish, validate, and test the machine learning methods using a cross-validation method. The experimental results, supported by statistical tests, demonstrate that the hybrid EO-LSSVR achieved the best prediction performance in terms of root mean square error (110.98), mean absolute percentage error (2.66%), mean absolute error (44.38), and coefficient of determination (0.958). The new machine learning models proposed in this study effectively overcome limitations inherent in theoretical and experimental studies and help increase prediction accuracy. Therefore, these models offer a promising tool for obtaining early and accurate estimates of structural durability, which are critical to scheduling and performing effective building maintenance.