Predicting Autogenous Shrinkage of Concrete Including Superabsorbent Polymers and Other Cementitious Ingredients Using Convolution-based Algorithms

Abstract

In this paper, the effectiveness of ensemble convolution-based deep learning models is evaluated for predicting autogenous shrinkage/swelling of cementitious materials. Various ensemble learning techniques are employed, including Simple Average Ensemble, Snapshot Ensemble, and Stacked Generalization, to develop predictive models. The models are trained and evaluated using performance metrics such as Root Mean Squared Error, Coefficient of Determination, Overall Index of model performance, Mean Absolute Error, and 95% Uncertainty. The results show that the integrated stacking model (ISM) outperforms other models in terms of predictive accuracy. Furthermore, the SHapley Additive exPlanation (SHAP) technique was used to interpret the ISM model. The analysis reveals that the most influential factors affecting shrinkage predictions include time, aggregate to cement ratio (A/C), superabsorbent polymer (SAP) content, water to binder ratio, cement content, water to cement ratio, and silica fume content. Also, the ISM model was compared with models developed previously by other researchers, namely, K-Nearest Neighbors (KNN), Random Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGB). With the lowest RMSE and MAE values, the ISM model has exceptional accuracy, demonstrating its capacity to create predictions that closely resemble observed values. Additionally, it has the highest coefficient of determination value, demonstrating its effectiveness in explaining a sizable percentage of the data variance. The Overall Index (OI) statistic shows that the ISM model performs exceptionally well, indicating that it captures more of the underlying information in the data. Additionally, it displays lower 95% confidence intervals, demonstrating greater assurance in its forecasts.