Gradient Boosting‐Based Modeling of Structural Performance in Hybrid Multi‐Tube Columns

This study presents an advanced predictive framework for estimating the ultimate confined strength ( f cc , u ) and ultimate strain ( ε cc , u ) of a hybrid multi‐tube concrete column (MTCC), an innovative structural system that combines a fiber‐reinforced polymer (FRP) outer tube, inner steel tubes, and void‐filled concrete. The study aims to improve the accuracy of predicting the structural performance of this system using machine learning (ML) techniques, particularly graded reinforcement models (GBMs). A database of 283 specimens generated from laboratory experiments and finite element simulations was used, incorporating the geometric and material properties affecting performance. Four different models were developed: random graded reinforcement (stochastic gradient boosting (SGB)), heavy (XGBoost (XGB)), light (LightGBM (LGB)), and class (CatBoost (CGB)), and their performance was optimized using Bayesian parameter tuning techniques. The evaluation results showed that the SGB model was the most accurate, achieving the highest coefficients of determination ( R 2 = 0.994 for f cc , u and 0.946 for ε cc , u ) and the lowest root mean square error (RMSE). SHapley Additive exPlanations (SHAP) analysis was used to interpret the model logic, revealing that the concrete compressive strength ( f ′ c ) and FRP layer thickness ( t f ) had the greatest influence on f cc , u , while the FRP elastic modulus ( E f ) and steel pipe yield strength ( f ys ) were the most important factors in determining ε cc , u . To enhance practicality, an interactive graphical interface was developed that allows engineers and researchers to make accurate predictions in an easy and efficient manner. The results confirm the effectiveness of ML models, particularly gradient boosting, in supporting design and analysis decisions in advanced engineering applications.