Experimental and machine learning evaluation of crumb rubber concrete strength at elevated temperatures.

This study addresses the limited understanding of the strength performance of crumb rubber concrete when exposed to elevated temperatures, particularly the lack of integrated experimental evidence and reliable predictive models for structural applications. The difficulty in quantifying strength degradation and identifying the dominant influencing parameters limits the wider use of crumb rubber concrete in fire related environments. To address this issue, an integrated experimental and machine learning framework is proposed to evaluate and predict the compressive and tensile strength of crumb rubber concrete subjected to high temperature exposure. The experimental program involved replacing fine aggregate with fine crumb rubber at different replacement levels, followed by mechanical testing after exposure to elevated temperatures. The experimental results indicate that increasing the crumb rubber replacement leads to a reduction in both compressive and tensile strength, which is attributed to changes in the concrete microstructure and weakened bonding between the rubber particles and cement matrix, with additional degradation observed at higher temperatures. To complement the experimental investigation, advanced machine learning models including XGBoost, Light Gradient Boosting, and Multilayer Perceptron were developed and achieved high prediction accuracy with coefficient of determination values exceeding 0.90, whereas simpler models such as K Nearest Neighbors and Decision Trees showed lower predictive performance. Model interpretability using SHAP analysis revealed that crumb rubber content, exposure temperature, and curing age are the most influential parameters governing strength prediction. The outcomes of this study provide a reliable predictive framework for optimizing crumb rubber concrete performance under elevated temperature conditions, while future research should consider additional durability parameters, extended temperature ranges, and larger datasets to further improve model generalization and practical applicability in sustainable construction.