PSI - Issue 70

Sahil Sehrawat et al. / Procedia Structural Integrity 70 (2025) 394–400

399

KFCV, R2, MAE, and RMSE for DT and RF models in table 3. the DT model's lowest, mean, and highest MAE values: 4.14, 11.59, and 23.11. shows the RF model's MAE varied from 2.59 to 13.28. The DT and RF models had 14.66 and 9.63 RMSEs. The mean R2 values for DT and RF were 0.52 and 0.62. The RF model with the lowest error and greatest R2 predicts concrete compressive strength best. MAE, RMSE, and Table 3 shows R2 k- fold analysis findings for each model.

Table 3. Results obtained from KFCV for each and every model that was used.

DT

K-Fold

RF

MAE 11.67 23.11 13.44 12.61

RMSE

R2

MAE

RMSE

R2

1 2 3 4 5 6 7 8 9

14.28 26.44 20.82 15.43

0.11 0.87 0.75 0.21 0.79 0.88 0.33 0.47 0.17 0.64

9.26

11.43 14.38

0.95 0.70 0.62 0.77 0.34 0.78 0.55 0.45 0.39 0.66

13.28

9.37

3.70

12.75

15.74 11.38

6.00 4.14 6.92

6.26 7.10 9.62

6.40 5.27 9.84 2.59

8.19 8.60 6.94 7.55 8.38

12.10 15.68 10.25

13.81 22.15 10.69

13.28

10

8.75

4. Conclusion Geopolymer concrete compressive strength was assessed using supervised machine learning (SML) techniques, both individual and ensemble. We predicted outcomes using random forest (RF) ensemble methods and decision tree (DT) individual approaches. This research found: SML approaches outperformed individual strategies in predicting GPC CS, with the BR model being most accurate. R2 values for Random Forest (RF) and DT models were 0.95 and 0.88. All models performed well with little data variation. In the sensitivity study, fly ash, GGBS, NaOH molarity, water-to-solids ratio, sand, gravel (10-20 mm), NaOH, gravel (4/10 mm), and Na2SiO3 accounted for 26.4%, 14.7%, 13.1%, 11.6%, 9.5%, 7.5%, 6.5%, 5.8%. This project estimates building material strengths fast and cheaply. Sustainable construction practises will boost GPC use in building projects. Laboratory experiments should improve data. Assess model performance for geopolymer, concrete, aggregate, and admixtures. Acknowledgements The authors would like to acknowledge SRM University, Sonepat, Haryana, India, for providing the technical and financial support necessary to conduct all laboratory activities. Special appreciation is extended to the undergraduate students of the Department of Civil Engineering and the Department of Computer Science for their technical assistance. The authors also express their gratitude to the HOD, Dr. Abhay Kumar Choubay and Dr. R. Mohanraj from the Department of Civil Engineering for their contributions to the study on the Prediction of Geopolymer Composite Strength Using Soft Computing Techniques.

References

Ahmad, J., Khan, M. and Ali, A., 2021e. Predictive modeling of mechanical properties of concrete incorporating waste using ensemble machine learning techniques. Construction and Building Materials, 300, p.124284.