Modeling and analysis of creep in concrete containing supplementary cementitious materials based on machine learning

This paper used machine learning to model the prediction of creep and the analysis of characteristic factors for concrete containing supplementary cementitious materials (SCM). First, a creep database covering thirteen input parameters and one output parameter was developed. Then, based on this database, Linear Regression (LR), Support Vector Regression (SVR), Random Forest (RF) and Extreme Gradient Boosting (XGB) models were used to establish creep compliance models. Finally, the sensitivity analysis of each input parameter based on the shapley additive explanation (SHAP) model was performed to investigate the influence of each parameter on the prediction results of creep. The results show that the extreme gradient boosting (XGB) model can predict the creep compliance of SCM concrete very well. The creep compliance is the result of the combined effect of numerous positively and negatively correlated characteristic parameters. Among the many characteristic pa-rameters, the stress-to-strength ratio is the most important factor affecting the prediction of creep. The creep compliance of concrete is positively correlated with cement content, while it is negatively correlated with silica fume (SF) content, and shows a negative and then positive change with slag and fly ash (FA) content. The creep tendency is smaller when the slag content is about 25% or the FA content is about 20%. Meanwhile, the interaction interval between different SCM admixtures and cement content differed in concrete creep prediction.