Development of extreme gradient boosting model for prediction of punching shear resistance of R/C interior slabs

Abstract

This paper aims to present the application of extreme gradient boosting (XGBoost) to the prediction of the punching shear resistance of reinforced concrete (R/C) interior slabs without shear reinforcement. For the training and testing of the XGBoost model, which was developed using the XGBoost 1.1.1 package, 497 experimental data of interior slab-column connections were collected from the literature. The input variables were the column section dimension, slab effective depth, concrete compressive strength, steel yield strength, and reinforcement ratio at the top and bottom of the slab. The targeted output variable was the punching shear strength. First, the developed XGBoost model was compared with two other machine learning (ML) models that incorporate artificial neural network (ANN) and random forest (RF). All three ML models could reliably estimate the punching shear resistance of the considered type of R/C slabs, but the XGBoost model generally achieved the best prediction. Second, the performance of the developed XGBoost model was compared to various design codes and empirical models. The XGBoost model presented the most accurate prediction among them with the coefficient of determination (R-2) for the testing dataset being equal to 0.9578. Third, the relative significance of input variables in the prediction of punching shear resistance was examined. The effective depth was shown to have the most significant role in the punching shear prediction. Finally, a graphical user interface based on the XGBoost model was created for preliminary estimation of the punching shear resistance of R/C interior slabs without shear reinforcement.