An optimization-based stacked ensemble regression approach to predict the compressive strength of self-compacting concrete

This research paper presents a study on predicting the compressive strength of self-compacting concrete (SCC) containing glass aggregate. A stacked ensemble approach was employed, which is a method of combining multiple models to improve the overall performance. The ensemble consisted of gradient boosting, extreme gradient boost, random forest, and K-nearest neighbors regressors as base learners, and linear regression as the glass aggregates (FGA), coarse aggregates, coarse glass aggregates (CGA), and superplasticizer were taken as input variables and compressive strength as output variables. The hyperparameters of the base learners were optimized using tree based pipeline optimization (TPOT). The ensemble's accuracy was evaluated using the K-fold cross-validation technique and statistical metrics. The performance of the stacked ensemble models is found to be better than other machine learning models. Permutation feature importance was used to determine the importance of the features in predicting compressive strength. The results demonstrate that the stacked ensemble approach with R2 = 0.9866, RMSE = 1.4730, and MAE = 1.0692 performed better than the individual base learners and the other machine learning models. The water-binder ratio has the highest impact on predicting the compressive strength of SCC containing glass aggregate.