Data-driven prediction of compressive strength in rice husk ash concrete using machine learning for sustainable construction

Abstract Rice husk ash (RHA) has the potential to partially replace ordinary Portland cement and increase the sustainability of concrete production, but optimizing the compressive strength of RHA concrete is challenging due to the complexity of interplay between mix-design and curing variables in the conventional laboratory approach which consumes a significant amount of time and resources. The present study is an attempt to propose an interpretable machine-learning model for predicting the compressive strength based on 9 input variables obtained from published experimental studies, such as RHA replacement, water-to-cement (W/C) ratio, curing time, aggregate size, superplasticizer dosage, cement content, fine aggregate content, coarse aggregate content and water content. The target variable is taken as the compressive strength. The four increasingly challenging validation scenarios included five regression algorithms: Linear Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). GridSearchCV using 5 fold cross validation was used to optimize Hyperparameters. SVM had the best performance in the final optimized comparison with R² = 0.875641, RMSE = 5.919637 MPa and MAE = 4.813035 MPa, whereas RF showed almost similar performance with R 2 = 0.860683 and also lower MAPE value. Curing time and W/C ratio were found to be influential predictors based on SHAP, Partial Dependence Plots (PDP), and Individual Conditional Expectation (ICE) analysis. Since there are several mix-design variables which are physically coupled, these results on interpretability are considered to be associations of the variables rather than causal relationships. The prediction performance was reduced significantly in the most difficult validation situation where both the cement content and aggregate size were not the same as those used during model training. Hence the present framework is meant to be used as a tool for decision making within the set limits of parameters studied and needs independent experimental verification before being adopted in engineering.