Metaheuristic-based machine learning approaches of compressive strength forecasting of steel fiber reinforced concrete with SHapley Additive exPlanations

Steel-fiber-reinforced concrete (SFRC) has proven to be a practical and effective alternative to traditional concrete. It offers improved post-cracking performance, increased fracture resistance, and efficient stress transfer by incorporating steel fibers into the concrete mix. Machine learning (ML) techniques are widely used to accurately estimate concrete qualities, thus reducing time and costs. This research focuses on a novel hybrid machine learning model that such as Gaussian process regression with genetic algorithm (GPR-GA), Random Forest with genetic algorithm (RF-GA), and extreme gradient boosting with genetic algorithm (XGB-GA) to predict the compressive strength of SFRC. The proposed hybrid ML models demonstrate superior performance with an R² value over 0.92 at both the training and testing stages. Specifically, the R² value was found to be 0.9804 for GPR-GA, 0.9204 for RF-GA, and 0.9706 for XGB-GA. The RMSE values for GPR-GA, RF-GA, and XGB-GA were 8.38, 16.88, and 16.88, respectively. The overall GPR-GA model outperformed compare to the other two hybrid machine learning models, XGB-GA and RF-GA. Additionally, the SHAP method was utilized to assess the influence of input variables on the predicted compressive strength (CS) of SRFC. From the SHAP analysis, it was found that the temperature and diameter input parameters had the highest influence on the predicted CS compared to other input features. The proposed hybrid ML models provide builders and designers with a flexible and effective tool for analyzing characteristics and making accurate forecasts of the compressive strength of SFRC under high temperatures in building applications.