Compressive strength prediction of sustainable concrete incorporating rice husk ash (RHA) using hybrid machine learning algorithms and parametric analyses

The construction industry is making efforts to reduce the environmental impact of cement production in concrete by incorporating alternative and supplementary cementitious materials, as well as lowering carbon emissions. One such material that has gained popularity in this context is rice husk ash (RHA) due to its pozzolanic reactions. This study aims to forecast the compressive strength (CS) of RHA-based concrete (RBC) by examining the effects of several factors such as cement, RHA content, curing age, water usage, aggregate amount, and superplasticizer content. To accomplish this, the study collected and analyzed data from literature, resulting in a dataset of 1404 observations. Several machine learning (ML) models, such as light gradient boosting (LGB), extreme gradient boosting (XGB), and random forest (RF), as well as hybrid machine learning (HML) approaches like XGB-LGB and XGB-RF were employed to thoroughly analyze these parameters and assess their impact on strength. The dataset was split into training and testing groups, and statistical analyses were performed to determine the relationships between the input parameters and CS. Moreover, the performance of all the models was evaluated using various statistical evaluation criteria, including mean absolute percentage error (MAPE), coefficient of efficiency (CE), root mean square error (RMSE), and coefficient of determination (R2). The hybrid XGB-LGB model was found to have higher precision (R2 = 0.95, and RMSE = 5.255 MPa) as compared to other models. SHAP (SHapley Additive exPlanations) analysis revealed that cement, RHA, and superplasticizer had a positive effect on strength. Overall, the study's findings suggest that the hybrid XGB-LGB model with the identified input parameters can be used to accurately predict the CS of RBC. The application of such technologies in the construction sector can facilitate the rapid and low-cost identification of material qualities and the impact of input parameters.