The compressive strength of crumb rubberized concrete (CRC) is vital for its structural performance and sustainability. This study integrates XGBoost machine learning with advanced regression and visualization techniques to optimize CRC strength across multiple cement grades. It presents one of the first comprehensive analyses combining feature selection, in-depth outlier detection, uncertainty quantification, and sensitivity analysis across 10–18 input variables and 531 samples. This study emphasizes the significance of robust feature selection, with the OG18 model achieving the highest accuracy (R2=0.979, RMSE=2.63 MPa) and the lowest prediction uncertainty. In contrast, the SSV10 model prioritizes speed, resulting in lower precision and increased prediction uncertainty. Notably, variance inflation factor (VIF) thresholds (5 or 10) were found to be misleading; models with VIF values up to 45 outperformed those adhering to conventional cutoffs. Additionally, the findings indicate that outliers may not represent errors but rather critical data points that enhance model predictions. Furthermore, shapley additive explanations (SHAP) values, permutation importance, and partial dependence plots consistently identified crumb rubber (CR) content, w/b ratio and curing age as dominant strength predictors. Conceptual strength profiles and weight-volume rubber interaction were developed to assist in mix design optimization. Findings show that using <135 kg/m3 (∼30%) of CR with G52.5 cement can replace up to 379 kg/m3 of sand while achieving structural-grade strength, whereas G32.5 is better suited for nonload-bearing applications. This study offers a novel, data-driven framework for sustainable CRC design, bridging key knowledge gaps in ML-based material modeling. These outcomes reinforce confidence in adopting practices that align with SDGs 8, 11, and 12. Looking ahead, future model enhancements are proposed to address the identified limitations and further refine predictive accuracy.
https://doi.org/10.1061/JMCEE7.MTENG-2071