An Innovative Machine Learning Model to Formulate the Bearing Capacity of Concrete-Filled Steel Tube Column

Concrete-filled steel tube (CFST) as the high-tech composite members utilized as a main load-carrying element in high-rise buildings' construction. CFST element load capacity is considered one of the most crucial and challenging engineering parameters for designing columns structurally and economically for steel-concrete composite. Because of the designing complexity of theoretically simulation and serviceability limits, this paper attempted to overcome the engineering problem using a machine learning (ML) methods. To do so, numerous efficient ML modeling called multivariate adaptive regression spline (MARS), M5p model tree (M5p), extreme learning machine (ELM), and random forest (RF) are implemented to propose a new auto-estimated and interpretable model. Through extensive literature, including 1305 (circular column) and 1003 (rectangular column) subjected to concentric axial force, data-intelligence models are developed. The developed models were compared with corresponding values computed by design code provisions, including Eurocode 4, LRFD, AISC 360-16, AS5100, ACI 318-14, and empirical equations extracted. The statistical metrics present that the proposed MARS models (r = 0.990, RMSE = 361.32 KN, WI = 0.995, and PMARE = 14.078% (circular column)) and (r = 0.974, RMSE = 494.94 KN, WI = 0.984, and PMARE = 11.238% (rectangular column)) boosted the performance of the simulation of the CFTS column compare to other models and design codes. In addition, global sensitivity analysis was performed using SOBOL methods to evaluate effective parameters. The explicit simulation model of the CFST columns is satisfied with the parametric study and shows the ability to perform the modeling and the cost-effective benefits of the information approach.