Metaheuristic-based prediction of flexural strength and dual-objective optimization for circular concrete-filled steel tubes

Concrete-filled steel tube (CFST) structures are extensively used in high-rise buildings, bridges, and subway stations due to their superior mechanical properties. However, accurately predicting their flexural strength capacity (M_u) remains challenging, as traditional empirical models and design codes fail to capture the complex nonlinear interactions between steel and concrete, leading to significant errors. To address this limitation, an advanced Jellyfish Search Optimizer (JSO)-enhanced XGBoost model is proposed, significantly improving M_u prediction accuracy. The model achieves R² = 0.9986 and RMSE = 5.81 kNm on the test set, demonstrating outstanding predictive performance. Moreover, Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDP) enhance model interpretability by revealing feature importance and interactions. Additionally, comparisons with AISC, Eurocode 4, and Han's Equation confirm its superiority. Furthermore, a user-friendly graphical user interface (GUI) is developed to enable real-time M_u predictions, facilitating practical engineering applications. Finally, the Non-Dominated Genetic Sorting algorithm II (NSGA-II) was employed for dual-objective optimization, successfully balancing M_u and material cost (C_total) of CFST.