Abstract
The increasing demand for sustainable construction materials has led to the incorporation of Palm Oil Fuel Ash (POFA) into concrete to reduce cement consumption and lower CO₂ emissions. However, predicting the compressive strength (CS) of POFA-based concrete remains challenging due to the variability of input factors. This study addresses this issue by applying advanced machine learning models to forecast the CS of POFA-incorporated concrete. A dataset of 407 samples was collected, including six input parameters: cement content, POFA dosage, water-to-binder ratio, aggregate ratio, superplasticizer content, and curing age. The dataset was divided into 70% for training and 30% for testing. The models evaluated include Hybrid XGB-LGBM, ANN, Bagging, LSSVM, GEP, XGB and LGBM. The performance of these models was assessed using key metrics, the coefficient of determination (R2), root mean square error (RMSE), normalized root means square error (NRMSE), mean absolute error (MAE) and Willmott index (d). The Hybrid XGB-LGBM model achieved the maximum R2 of 0.976 and the lowest RMSE, demonstrating superior accuracy, followed by the ANN model with an R2 of 0.968. SHAP analysis further validated the models by identifying the most impactful input factors, with the water-to-binder ratio emerging as the most influential. These predictive models offer the construction industry a reliable framework for evaluating POFA concrete, reducing the need for extensive experimental testing, and promoting the development of more eco-friendly, cost-effective building materials.