Development of machine learning methods to predict the compressive strength of fiber-reinforced self-compacting concrete and sensitivity analysis

Fiber-reinforced self-compacting concrete (FRSCC), a great combination of self-compacting concrete (SCC) and fiber, plays a vital role as a potential construction material. Improving the accuracy of FRSCC' performance prediction methods is critical and challenging to reduce costly experiments and time. Therefore, this study developed and assessed the performance of three machine learning models, including Decision tree, Light Gradient Boosting Machine, and Extreme Gradient Boosting (XGBoost), for predicting the compressive strength (CS) of FRSCC. The models were developed based on 387 data samples with 17 input parameters. Monte Carlo and K-fold cross-validation techniques were used to assess the models' generalizability and predictive performance. The results showed that the XGBoost model has the highest predictive performance and stability, with typical results R2 = 0.992, RMSE = 1.892 MPa, MAE = 1.438 MPa. The sensitivity analysis of the models indicated that cement, coarse aggregate, fine aggregate, water, and sample age significantly influence the CS of FRSCC with inconsistent order. Finally, XGBoost was the most accurate and reliable model based on the final architecture analysis.