An efficient prediction of punching shear strength in reinforced concrete slabs through boosting methods and metaheuristic algorithms
Erfan Khajavi, Amir Reza Taghavi Khanghah, Ali Javadzade Khiavi
Abstract
Safety and optimum designs in civil engineering applications depend on correctly estimating critical parameters such as punching shear strength (Pu) in concrete slabs . This work, therefore, studies improvement in prediction accuracy and generalization by applying five advanced boosting machine learning methods to overcome traditional empirical methods' drawbacks. Five state-of-the-art boosting models, namely, XGBoost Regression (XGBr), Histogram Gradient Boosting Regression (HGBr), Light Gradient Boosting Regression (LGBr), CatBoost, and AdaBoost, were conducted in modeling Pu prediction. K-fold cross-validation has been conducted to select only the best two models for further elaboration with state-of-the-art optimization algorithms, including Quadratic Interpolation Optimization (QIO), Red-tailed Hawk Algorithm (RTH), Giant Armadillo Optimization (GAO), and Greylag Goose Optimization (GGO). Some of the ensemble techniques were also used to enhance robustness and accuracy. Several sensitivity analysis methods were conducted to determine the effect of input parameters on model performance. Indeed, very important insight was provided about major contributing factors. The proposed techniques have certain significant advantages over others. It can process nonlinear relationships, provide a minimum prediction error, and recognize those parameters critical to punching shear. This paper emphasizes how integrating machine learning with optimization and sensitivity analysis can enable the development of state-of-the-art predictive modeling in structural engineering that is reliable and more efficient than traditional methods.