Predicting the crack width of the engineered cementitious materials via standard machine learning algorithms

This study aims to develop various machine learning (ML) models to investigate the self-healing capacity of engineered cementitious composites (ECC) and to evaluate the effect of input parameters (raw materials and crack width before the healing process) on output parameters (crack width after the healing process) via Shapley additive explanations (SHAP) analysis. The combination of the individual and ensemble ML models has been introduced to check and compare the accuracy level towards the prediction of crack width after the healing process to access the most suitable model for this application. The support vector machine (SVM) from the individual, while XGBoost (XGB) and random forest (RF) from ensemble ML models have been investigated for prediction purposes. As per the obtained results, the RF model outperforms both the SVM and XGB algorithms in forecasting the fracture width after the healing process for the selected ECC in concrete material. The performance indicator, such as the coefficient of determination (R2), was reported as 0.97 for RF, 0.96 for XGB, and 0.93 for the SVM model. Statistical results and k-fold cross-validation for the employed ML models also confirm their legitimacy. It was also noted from the result of the SHAP analysis that the significant contribution was for the crack width before healing towards the prediction of crack width after the healing process. Furthermore, the ML models can also be utilized to anticipate the healing ability of the other ECC, such as soda glass powder, marble powder, and bagasse ash.