Prediction of fresh and hardened properties of self-compacting concrete using ensemble soft learning techniques

In this study, the application of an ensemble machine learning (ML) model to estimate the fresh and hardened properties of self-compacting concrete (SCC) is studied. Total six input variables were considered that covers dosage of cement, fly ash, water–powder ratio, coarse aggregate, fine aggregate, and super-plasticizer, whereas slump flow, L-box ratio, V-funnel, and compressive strength (Fc28) were used as prediction variables. Total 3 machine learning algorithms have been used, such as gradient boosting (GBR), Adaboost, LightGBM, XGBoost, and CatBoost to forecast the fresh and hardened properties of SCC. The prediction performance of all the models for both fresh and hardened properties was compared using a testing dataset, and it was noticed that the XGBoost and CatBoost model exhibit more accurate prediction than the gradient boosting algorithm model. Analysis of the study reveals that XGBoost model has performed well in slump flow and V-funnel prediction by high correlation coefficient of 0.9941 and 0. 9981. Similarly, CatBoost model has performed efficient in predicting L-box and compressive strength by high correlation coefficient of 0.9926 and 0.9941. Moreover, a sensitivity analysis has also been performed to investigate the effect of input ingredients on all other output variables.