Abstract: The mix design of foam lightweight concrete exhibits complex nonlinear relationships with its mechanical properties, making high-precision prediction challenging for traditional empirical models. This study proposes an XGBoost hybrid intelligent prediction model, integrating optimization algorithms with ensemble learning for compressive strength forecasting. A multi-source database comprising 323 experimental data sets was constructed. Using foam lightweight concrete density, water-cement ratio, and sand-cement ratio as inputs for predicting compressive strength, a nonlinear mapping relationship between input parameters and compressive strength was established. Eighty percent of the sample data served as the model training set, with the remaining samples forming the test set. Unlike traditional single-model approaches, this study constructed a comparative validation system incorporating SVR, BPNN, and XGBoost. Model performance evaluation revealed that the XGBoost model achieved a coefficient of determination R²=0.922 and a mean squared error MSE=13.549 on the test set, significantly outperforming both the SVR (R²=0.902, MSE=17.435) and BPNN (R²=0.893, MSE=20.031) models. The proposed hybrid intelligent modeling approach overcomes the accuracy limitations of traditional empirical formulas, offering a novel technical pathway for the intelligent design and performance prediction of foam lightweight concrete materials. This holds significant engineering application value.