Abstract In structural engineering, accurately predicting the axial load capacity of concrete‐filled double‐skin tubular columns reinforced with fiber‐reinforced polymers is a vital challenge. This study addresses this by developing a single, unified predictive CatBoost model that generalizes across diverse fiber‐reinforced polymers (FRP) types, including carbon FRP, aramid FRP, glass FRP, and polyethylene terephthalate FRP. The model development involves fine‐tuning a CatBoost algorithm with three distinct optimization methods, including Salp Swarm Optimization, a hybrid Gray Wolf‐Whale Optimization algorithm, and the Moth Search algorithm. Using a dataset of 280 experimental results, the CatBoost models achieved excellent performance on the held‐out test data ( R 2 = 0.962 and 0.960), and outperformed seven other fully optimized machine learning (ML) models in a fair benchmark comparison (CatBoost default, gradient boosting regression, Extreme Gradient Boosting, Histogram Gradient Boosting Regressor, Light Gradient Boosting Machine, Random Forest, and Adaptive Boosting Machine). To establish the model's reliability for engineering practice, its black box nature was investigated. A SHapley Additive exPlanations analysis was used to confirm the model's interpretability, and a parametric study verified that its predictions are physically consistent. This research provides a validated predictive tool, distributed with a graphical user interface, to simplify the design process and support the practical application of ML in structural engineering.
Trung-Kien Nguyen (Thu,) studied this question.