Predicting the compressive strength of fiber-reinforced self-consolidating concrete using a hybrid machine learning approach

Fiber-reinforced self-consolidating concrete (FR-SCC) combines the advantageous characteristics of self-compacting concrete with fiber reinforcement, providing a versatile solution for contemporary construction. However, due to its complexity and the scarcity of available data, the strength prediction techniques of FR-SCC are still in their early stages. To get around this limitation, research was done to create an optimal machine learning algorithm for predicting the compressive strength (CS) of FR-SCC. This work aims to precisely forecast the CS of FR-SCC by optimizing the parameters and structure of a Levenberg-Marquardt back propagation Artificial Neural Network (LMBP-ANN) model using K-fold cross-validation. One hundred twenty-three experimental data on FR-SCC from available literature was used to create the dataset. Several validation metrics, including coefficient of determination (R2), mean absolute error (MAE), and root mean square error (RMSE) were employed to validate the models. Essential features that significantly impact the complex behavior of FR-SCC were found and incorporated into the model using multivariate analysis, Pearson correlation chart, and feature selection. The results show that K-fold cross-validation reduced training and testing errors by 22.2% and 18.3%. Consequently, an R2 value of 0.9343 was achieved, which validated the model's accuracy. SHAP analysis was also conducted in order to interpret the contribution of different features to the strength of FR-SCC. The most impactful feature was coarse aggregate, followed by curing age, superplasticizer, fly ash, and fiber content. The current work's findings might aid in precisely predicting the FR-SCC and the ANN network's design optimization procedure.