Advanced machine learning approaches for predicting compressive and flexural strength of carbon nanotube–reinforced cement composites: a comparative study and model interpretability analysis

Abstract The accurate prediction of the compressive strength (CS) and flexural strength (FS) of carbon nanotube (CNT)-reinforced cement composites is critical for optimizing their performance in construction applications. However, traditional experimental investigations are time-consuming and resource-intensive, while existing predictive models struggle to capture the complex relationships between CNT incorporation and the resulting mechanical properties. Accordingly, this study aimed to develop a reliable and efficient predictive framework by employing advanced machine learning (ML) techniques, including support vector regression (SVR), SVR-Bagging, SVR-Boosting, random forests (RF), gradient boosting (GB), and decision trees (DT). The models were trained and validated using available experimental data, and their interpretability was enhanced through SHapley Additive exPlanations (SHAP) and partial dependence plots (PDPs), which were applied to optimize the mix design by identifying the most influential parameters affecting CS and FS. Among the developed models, GB demonstrated the highest prediction accuracy, achieving R 2 values of 0.993 for CS and 0.987 for FS, outperforming RF (0.990 and 0.971) and other ML approaches. In addition, an interactive user interface was created to facilitate practical applications, allowing engineers to predict CS and FS values rapidly without physical testing. This study contributes to the existing literature by providing a comprehensive, interpretable, and practically applicable ML-based framework that advances predictive modeling and mix design optimization for CNT-reinforced cement composites.