Predicting the mechanical properties of concrete from waste glass through the use of boost and random forest

This study explores the partial replacement of cement with waste glass powder (WGP) in M25 and M30 grade concrete, ranging from 0% to 20% in 2.5% increments. The mechanical properties assessed include compressive, flexural, and split tensile strengths, as well as predictive modeling using Boost and Random Forest. The optimum performance was observed at 10% WGP replacement, where compressive strength reached 30 MPa for M25 and 35 MPa for M30 after 28 days. Strength declined beyond this point due to cement dilution and possible alkali-silica reaction effects. Flexural and split tensile strengths followed a similar trend, with maximum values recorded at 20% replacement. Slump values dropped from 75 mm to 55 mm for M25 and 53 mm for M30, mainly due to the angular shape and non-absorbent nature of WGP, which reduces workability. A small density reduction of about 2.5% was also noted. Linear regression showed strong correlations between curing time and strength, with R² values of 0.9631 (3 days), 0.8352 (7 days), and 0.8808 (28 days). Negative R² values indicate the XGBoost performed worse than predicting the mean. This behavior was mostly observed for early-age data and is likely due to limited variability and sample size. It also highlighted that Random Forest consistently outperformed XGBoost, especially for 28-day strengths, with positive and moderate-to-high R² scores.