Nature-inspired optimization of weighted-feature ensemble model to predict the deflection of corroded reinforced concrete beam

Steel corrosion is a significant concern for the structural integrity of reinforced concrete (RC) beams, leading to a decline in durability throughout a building's lifespan. Assessing the deflection of corroded RC beams is essential for maintaining structural safety and longevity. This study introduces a novel nature-inspired weighted-feature ensemble model (NiwE) that integrates two distinct machine learning algorithms: least squares support vector regression and radial basis function neural network, both optimized by beluga whale optimization to predict the deflection of corroded RC beams. The primary objective is to enhance the predictive accuracy of deflection measurements, offering a robust tool for evaluating structural performance under corrosioninduced damage. The model was developed and validated using a comprehensive dataset comprising 180 samples collected from aging residential buildings in southern Vietnam. A crossvalidation approach was employed in this process to ensure a robust and reliable evaluation of the models. The results show that the NiwE model achieved superior predictive accuracy compared to existing methods, with an RMSE of 1.739 mm, MAE of 1.115 mm, and R2 of 0.926 in testing. Remarkably, the NiwE model was able to determine weight values that facilitate the discovery of new feature combinations, achieving significantly higher accuracy than existing models. The proposed model offers civil engineers a valuable tool for estimating deflection in corroded RC beams and opens new pathways for optimizing building structures, developing early warning systems, and implementing timely maintenance measures to ensure safety and resilience.