Enhanced XGBoost algorithm with multi-objective optimization for blast-induced response forecasting of RC slabs

Amid increasingly frequent military conflicts and explosion events, accurately predicting the dynamic response of reinforced concrete (RC) slabs, key load-bearing components in building structures, is essential for understanding blast-induced damage and enhancing structural protection. However, current approaches predominantly rely on experimental tests, finite element (FE) simulations, and conventional machine learning (ML) techniques, which are often costly, inefficient, narrowly applicable, and insufficiently accurate. To overcome these challenges, this study aims to optimize ML models, refine architectural designs, and improve model interpretability. A comprehensive dataset comprising 489 samples was constructed by integrating experimental and simulation data from existing literature, incorporating 15 input features and one target variable. Based on this dataset, a novel method, termed MOPSO-TXGBoost, was proposed. Building on XGBoost as a baseline, the method employs multi-objective particle swarm optimization (MOPSO) for hyperparameter tuning, introduces a tri-stream stacking architecture to enhance feature representation, and trains three distinct models to improve generalization performance. A weighted fusion strategy is employed to further enhance the accuracy of prediction. Additionally, a model comprehensive evaluation (MCE) index is introduced, which integrates error metrics and fitting performance to facilitate systematic model assessment. Experimental results indicate that, compared with the baseline XGBoost model, the proposed approach reduces prediction error by 61.4% and increases the coefficient of determination (R²) by 0.217. Moreover, it outperforms several mainstream machine learning (ML) algorithms. The findings of this study advance ML-based blast damage prediction and provide theoretical support for safety assessment and protection optimization of RC slab structures.