Data-driven strategy for evaluating the response of eco-friendly concrete at elevated temperatures for fire resistance construction

Abstract

Concrete strength degrades drastically due to high temperatures (HT) destroying its components. Using supplementary cementitious materials (SCMs) in concrete and achieving the necessary compressive strength (CS) at HT is challenging and time-consuming. However, supervised machine learning (ML) approaches may be used to anticipate the expected outcome accurately, saving researchers time, effort, and money on tests. This research makes use of a decision tree (DT), Adaboost (ADB), and bagging regressor (BR) approach, to predict the CS of concrete at HT. Based on the findings of the coefficient of determination (R2), the BR model obtained the greatest value of R2 (0.92), followed by the ADB model (0.90) and the DT model (0.87). In addition, statistical analysis and k-fold cross-validation were carried out as part of the validation process for the models. According to these findings, the BR model had much lower MAE, RMSE, and RMSLE values than the ADB and DT models. Additionally, the Taylor diagram demonstrated that the values generated by the BR model were far closer to actual values than those generated by other models. This indicated that the BR model fared better than the ADB and DT models. In addition, sensitivity analysis revealed that temperature accounted for 66.3 % of the outcome, cement contributed 7.6 %, and coarse aggregates 8.4 %. Thus, ML models have been shown to be cost-effective and efficient in predicting SCM concrete CS at HTs.