Utilizing an enhanced machine learning approach for geopolymer concrete analysis

Abstract

ABSTRACT The flyash-based geopolymer concrete has been extensively utilized to minimize carbon emissions for over 15 years. The compressive strength (CS), split tensile strength (STS), and flexural strength (FS) of concrete determine its mechanical characteristics. Laboratory strength testing and analytical methods is costly and time-consuming, which struggle to predict complicated concrete combination CS, STS, and FS. ML models accurately predict CS, STS, FS. To predict the CS, STS, and FS of GPC concrete, five ML techniques (LR, SVM, GPR, EL, and ANN) were adopted. Taylor diagrams and error graphs adopted to represents model performances. The GPR outperforms other models in CS predictions by having the lowest MAE, RMSE, MAPE, and R-values. GPR continues to outperform in STS predictions, with low MAE, RMSE, and MAPE, as well as high R-values. GPR outperforms FS predictions with low MAE, RMSE, MAPE, and high R-values. SVM is excellent since it has a low MAE and RMSE. SVM, GPR, and ANN optimisations keep them competitive. GPR consistently gets the lowest MAPE values across all three tests, indicating that it is a strong model contender. The created model can forecast the mechanical characteristics of geopolymer concrete with ease and better reproducibility.