An intelligent approach for predicting strength of roadbed foam lightweight concrete based on optimized XGBoost model

The mix proportion design of roadbed foamed lightweight concrete (FLC) and its mechanical properties exhibit complex nonlinear relationships, making it difficult for traditional empirical models to achieve high-precision predictions. This study proposes a hybrid intelligent prediction model based on the Newton-Raphson-based Optimization (NRBO) algorithm improving eXtreme Gradient Boosting (XGBoost), integrating optimization algorithms with ensemble learning for FLC compressive strength prediction. A multi-source database containing 323 sets of experimental data was constructed, with FLC density (ρ), water-to-cement ratio (W/C), and sand-to-cement ratio (S/C) as input parameters. A nonlinear mapping relationship between input parameters and compressive strength was thereby established. Eighty percent of the sample data were used as the training set, and the remaining samples served as the test set. This study constructed a comparative validation system including SVR, BPNN, and XGBoost, and employed the NRBO algorithm to perform global optimization of XGBoost’s hyperparameter space. Model performance evaluation results show that the NRBO-XGBoost model achieved a coefficient of determination ( R 2 ) of 0.966, mean squared error ( MSE ) of 7.032, and mean absolute percentage error ( MAPE ) of 18.6 % on the test set. This represents an approximately 5 % improvement in prediction accuracy compared to the XGBoost model and significantly outperforms the SVR ( R 2 =0.902) and BPNN (R 2 =0.893) models. Sensitivity analysis of model input parameters revealed that the density parameter contributes 60 % to strength prediction, providing a theoretical basis for mix proportion optimization. Engineering validation test data confirmed that the model’s prediction errors were controlled within 10 %. The hybrid intelligent modeling method proposed in this study breaks through the accuracy limitations of traditional empirical formulas, offering a new technical pathway for intelligent design and performance prediction of FLC materials.