This study investigates the application of machine learning algorithms in predicting the compressive strength of concrete, a crucial performance parameter in the construction industry. Traditional concrete design methods rely heavily on extensive laboratory testing, making them costly and time-consuming. The research aims to develop more efficient predictive models to optimize material usage while maintaining structural integrity. The study employs advanced machine learning models, including CatBoost, XGBoost, and LightGBM, to estimate concrete strength based on varying proportions of cement, blast furnace slag (BFS), fly ash, and aggregates. To enhance model interpretability, Explainable Artificial Intelligence (XAI) techniques, such as SHapley Additive exPlanations (SHAP) and permutation-based characterization, were utilized to assess the influence of individual components on compressive strength. The results confirm that machine learning models effectively capture complex nonlinear relationships between concrete mix components and compressive strength. These models outperform conventional prediction approaches in terms of accuracy, providing a more reliable and efficient alternative for estimating concrete strength.
Deb et al. (Tue,) studied this question.