Self compacting concrete with recycled aggregate compressive strength prediction based on gradient boosting regression tree with Bayesian optimization hybrid model

Abstract

Self-compacting concrete (SCC) is a special type of concrete that is used in applications requiring high workability, such as in densely reinforced or complex formwork situations. The estimation of 28-day compressive strength for this type is usually made by costly and time-consuming laboratory tests. The problem becomes even more complex when recycled aggregates are added to the mixture to promote eco-friendly and sustainable construction practices. In our research we presented a new hybrid model, GBRT, that was integrated with Bayesian Optimization. This model is able to accurately and efficiently estimate the compressive strength of SCC containing recycled aggregates. We evaluated the model using well-known performance metrics such as RMSE, MAE, and. The performance of the model gave us, on average, an RMSE of 6.000, MAE of 3.968, and of 0.806 in five-fold cross-validation, which emphasized its strong predictive capability and potential as a cost-effective alternative to conventional laboratory testing. The model was also compared with single learner models such as SVR and KNN in order to demonstrate the superiority of the hybrid approach in terms of prediction accuracy and robustness. Our hybrid model surpassed the two previously mentioned models when testing their performance on the test data. Since our model works as a black-box model, a novel explaining machine learning technique named SHAP (Shapley Additive Explanations) was employed to determine which predictors have the most importance and how they trend. The developed model is an accurate, fast, and economical substitute for predicting 28-day compressive strength of self-compacting concrete with recycled aggregates. Finally, the model is converted into an easy-to-use graphical interface that provides civil engineers and practitioners with a useful decision-support tool for mix design optimization and quality control in real-life construction projects.