Integrating PCA and XGBoost for predicting UACLC of steel-reinforced concrete-filled square steel tubular columns at elevated temperatures

The residual strength and stability of Steel-Reinforced Concrete-Filled Square Steel Tubular (SRCFSST) columns under elevated temperatures are critical for fire safety engineering. This study presents a novel predictive framework utilizing a hybrid Principal Component Analysis and Extreme Gradient Boosting (PCA-XGB) model to estimate the ultimate axial compressive load capacity (UACLC) of SRCFSST columns. A dataset comprising 135 entries with eleven input parameters—encompassing geometric, material, and thermal characteristics—was pre-processed using PCA to address multicollinearity, reducing dimensionality while preserving variance. The PCA analysis identified that the first five principal components captured the most significant variability, with PC1 and PC3 showing the strongest correlations with UACLC, emphasizing their critical role in predicting structural behavior under extreme conditions. The PCA-XGB model, refined through Bayesian Optimization (BO), Tree-Structured Parzen Estimator (TPE), and advanced metaheuristic algorithms such as Whale Optimization and Directed Bee Colony Optimization, achieved exceptional predictive accuracy, with testing R 2 values reaching 0.947. Model performance metrics, including Root Mean Square Error (RMSE) and Weighted Mean Absolute Percentage Error (WMAPE), confirmed the robustness and reliability of the framework. SHapley Additive exPlanations (SHAP) analysis revealed that geometric parameters and concrete compressive strength were the most significant contributors to UACLC prediction. The proposed PCA-XGB model offers significant practical implications by providing engineers and designers with a reliable tool for optimizing fire-resistant structural designs, enhancing the resilience of critical infrastructure, and facilitating informed decision-making in high-temperature scenarios. The proposed PCA-XGB model provides a scalable, accurate, and interpretable tool for designing fire-resistant structures, bridging key gaps in real-time structural monitoring under extreme conditions.