Machine learning based prediction model for the compressive strength of fly ash reinforced concrete: an exploration of varying cement replacements and water-cement ratios

Abstract

This research explores the effect of fly ash replacement (0–50%) and various water-cement (W/C) ratios on the compressive strength of concrete. Experiments were conducted to evaluate compressive strength at different curing times (7, 28, 90, and 120 days) and W/C ratios (0.35, 0.45, 0.50). The results indicate that fly ash replacement reduces early-age compressive strength, with 50% fly ash mixes achieving around 12 MPa at 7 days compared to over 30 MPa for 0% fly ash, due to slower pozzolanic reactions. SVM-RBF, Random Forest, XGBoost, and Linear Regression based prediction models of compressive strength were developed. The performance of models was assessed using key performance metrics like MAE, MSE, RMSE, MSLE, RMSLE, R², MAPE, Willmott’s Index of Agreement, Mielke & Berry Index, and Legates & McCabe’s Index alongside Taylor diagrams, which revealed that SVM-RBF was the most reliable model, providing the best accuracy in both training (R² = 0.991) and testing phases (R² = 0.958). Sensitivity analysis indicated that curing days and water-cement ratio were the most influential factors on compressive strength, with curing days showing the highest normalized sensitivity index. Monotonicity analysis revealed optimal ranges for fly ash (~ 20%) and W/C ratio (~ 0.40) for maximizing compressive strength, with strength diminishing beyond these values due to increased porosity. The findings underscore the significance of fly ash content and W/C ratio in optimizing concrete strength, with machine learning providing valuable insights into predicting and understanding the behavior of concrete mixtures.