Prediction of compressive strength and tensile strain of engineered cementitious composite using machine learning

This research extensively used different progressive machine learning (ML) techniques to predict the compressive strength (CS) and tensile strain (TSt) of engineered cementitious composites (ECC) with 14 input variables and six algorithms. Specifically, random forest (RF), support vector machine, extreme gradient boosting (XGBoost), light gradient boosting machine, categorical gradient boosting (CatBoost), and natural gradient boosting techniques were used in the present study, to understand mechanical properties of ECC meanwhile these properties are crucial for design codes and developing new reliable models for mixtures. The discrepancy between the ML technique and specific ECC expected outputs is novel in this study and will aid researchers in better understanding of ECC features. To estimate the CS and TSt of the ECC, 2535 and 1469 input data points, respectively, were incorporated based on the material ratio, W/B, and different properties of the fibers. In addition, hyperparameter optimization techniques have also been used in ML to improve over fitting and make the model more accurate and robust. Moreover, an error analysis was highlighted between the actual and predicted CS and TSt of the ECC with each ML technique. Also, the significance and influence of the variable inputs that affect the CS and TSt were explained using the Shapley additive explanation (SHAP) approach. Among all approaches, CatBoost and XGBoost predicted the CS and TSt of ECC with greater accuracy than other techniques in terms of the coefficient of determination (R²), mean square error, mean absolute error, root mean square error, and symmetric mean absolute percentage error. The training and testing R² values of CatBoost and XGBoost for predicting the CS and TSt of ECC were 0.96, 0.89, 0.89, and 0.76, respectively. SHAP analysis revealed that W/B and fiber elongation were the most significant input variables for the CS and TSt of the ECC.

Graphical Abstract