This study presents a novel investigation into the effect of crumb rubber aggregates (RA) particle size on the compressive strength (CS) of rubberised concrete (RC), utilising five tree-based ensemble machine learning (EML) models validated through rigorous laboratory experiments and innovative analytical evaluations. As one of the few EML modelling studies, this analysis covers 361 literature-based and 25 laboratory-based samples incorporating 15 and 7 input variables, respectively. The findings revealed that the categorical boost model outperformed other EML models, achieving impressive evaluation metrics confirmed by cross-validation and Taylor diagrams. Sensitivity analysis highlighted the superiority of accumulated local effects compared to other techniques, pinpointing fine RA (0∼0.6 mm) as a critical factor influencing CS, as supported by both EML models and laboratory experiments. Higher RA content and water/cement (w/c) ratios reduce CS, but increased w/c can offset strength loss, making it suitable for low-strength and lightweight concrete applications. Furthermore, the developed hybrid expo-linear model combining exponential and Bolomey-based components captures multi-factor interactions well, with ±5% accuracy, and outperforming previous models for reliable rapid RC strength prediction. Overall, adapted EML-driven and expo-linear models match the concrete strength theory, clarifying non-linear interactions to optimize RC mixes whilst preserving structural performance and enhancing environmental sustainability.
https://doi.org/10.1080/10298436.2026.2662960