Data-driven compressive strength prediction of steel fiber reinforced concrete (SFRC) subjected to elevated temperatures using stacked machine learning algorithms

Experimental studies using a substantial number of datasets can be avoided by employing efficient methods to predict the mechanical properties of construction materials. The correlation between the mechanical attributes and structural performance of these structures can be determined using an efficient mathematical model. In this study, a large data-rich framework is constructed with data from 307 experiments conducted between 2000 and 2022 and reported in the literature to predict the compressive strength (CS) of steel fiber-reinforced concrete (SFRC) subjected to high temperatures. The collected data are utilized for training the proposed models using the SciKit, Tree-based Pipeline Optimization Tool (TPOT), and AutoKeras libraries in Python, followed by hyperparameter tuning and k-fold cross-validation. After performing the feature selection analysis, several machine learning (ML) algorithms are developed and compared. Out of 7 different leaderboard combinations, the best stacked pipeline including support vector machine, random forest, gradient boosting machine, extra tree regression, and K-nearest neighbors, is found to provide the most accurate solution. In addition, the results obtained using the stacked ML are compared with those obtained using an artificial neural network algorithm. Moreover, the accuracy of each method is determined through a comparative study. The stacked ML pipeline with optimum hyperparameters yields the highest accuracy (R2 = 0.92). The proposed stacked technique serves as an accurate and adaptable attribute evaluation tool for researchers to predict the CS of SFRCs subjected to elevated temperatures in construction applications.