Prediction of the Splitting Tensile Strength of Manufactured Sand Based High-Performance Concrete Using Explainable Machine Learning

A potential environmentally friendly building material is concrete containing manufactured sand. The utilization of manufactured sand in the production of high-performance concrete (HPC) has been gaining popularity. A commonly used mechanical parameter for the design of HPC structures is the splitting tensile strength. Due to the complexity, expense, and time-consuming nature of conducting tensile testing, numerous researchers are interested in creating a straightforward yet precise way to forecast the value of this feature. This study presents an effective application of machine learning models for predicting the tensile strength of HPC. Five advanced predictive algorithms, namely K-nearest neighbors, Adaptive Boosting (AdaBoost), Random Forest Regressor, Extra Tree Regressor (ETR), and Voting Regressor, were utilized. The performance of the developed ML models was then evaluated using distinct performance indexes. The evaluation shows that the ETR model’s estimated results are closer to the experimental results than the other four models. This suggests that the ETR model accurately estimates split tensile strength. Conversely, Shapley additive explanations (SHAP) offer comprehensive metrics for evaluating both the significance of features and the influence of a variable on a given prediction. It is noteworthy that the SHAP interpretations aligned with the typical tensile behaviour of the concrete, thus reinforcing the causal relationship between the machine learning predictions and the actual outcomes.