Artificial intelligence-based estimation of ultra-high-strength concrete's flexural property

Abstract

Advancement in Artificial Intelligence (AI) techniques and their applications in the construction industry, particularly for predicting mechanical properties of concrete, leads to conservation of efforts, time and cost. However, insufficient research has been done on the Ultra-high-strength concrete (UHSC). For this reason, this study aims to predict the UHSC flexural strength by applying sophisticated AI approaches. Ensembled machine learning techniques performed well compared to the individual decision tree (DT) model. In the current research, UHSC flexural strength is predicted by employing supervised Machine Learning (ML) approaches, i.e., DT-Bagging, DT-Gradient Boosting, DT-AdaBoost, and DT- XG Boost. Moreover, the model performance is assessed with the help of R2, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). In addition to that, the validation of model performance is also done by using the k-fold cross-validation technique. Higher R2 of 0.95 and lesser error (i.e., RMSE and MAE) values, in the case of the DT-Bagging model, depict improved model performance with respect to other applied ensemble methods. The assessment output shows that anticipated outcomes from proposed models, i.e., DT-Bagging, are much closer to actual results from experiments, which indicates the enhanced prediction of flexural strength for UHSC. Further, the SHapley Additive exPlanations (SHAP) analysis shows that steel fiber content has the highest positive influence on UHSC flexural strength.