Genetic algorithm hybridized with eXtreme gradient boosting to predict axial compressive capacity of CCFST columns

Abstract

The study aimed to propose a robust method for predicting the axial compressive capacity (Nu) of circular concrete-filled steel tube (CCFST) columns. For this purpose, a hybrid intelligent system was designed by fusing the eXtreme Gradient Boosting (XGB) model with the Genetic Algorithm (GA). The GA was utilized to determine an optimal set of XGB's hyperparameters to improve accuracy. A dataset of 509 tests from available literature was compiled for training and testing phases, which covered five input variables considering geometric and material properties of CCFST columns. The objective predictive model (GA-XGB) was validated against a benchmark model, namely GA-ANN, an artificial neural network (ANN) model optimized by the same metaheuristic algorithm. The simulation results in terms of error range and statistical indices indicated that the GA-XGB was superior to the GA-ANN model and other established mathematical approaches in predicting Nu of CCFST columns. The relative importance of individual features was further investigated to find the most significant input variable. Finally, a web app has been developed to make the proposed GA-XGB model applicable to practical design. The results confirmed that GA-XGB can be used in the design and behavior prediction of CCFST columns as a reliable and precise tool.