Using machine learning approaches for predicting the compressive strength of ultra-high-performance concrete with SHAP analysis

Ultra-high-performance concrete (UHPC) is a special construction material that is renowned for its exceptional mechanical properties and durability. Machine Learning (ML) methodologies have become essential methods for predicting the compressive strength (CS) of UHPC and identifying critical independent variables that are essential for optimal mix design. This study predicted the CS of UHPC using three ML models: artificial neural networks, M5P, and random forest. Various measures, including R, RMSE, RAE, MAE, and RRSE were employed to assess the performance of these models. A total of 810 observations were gathered, with 80% designated for training and the remaining 20% for testing. The RF model demonstrated superior performance compared to the other models, attaining a testing phase R-value of 0.98, MAE of 6.35, RMSE of 8.49, RRSE of 21.90%, and RAE of 21.21% through statistical analysis. The findings indicate that the variable “age” exerted the most significant influence on the model’s dependent variable, evidenced by a SHAP value of 23.06. ML techniques provide substantial benefits to the construction sector by facilitating faster and more accurate evaluations of material attributes.