Hybrid machine learning approach for predicting compressive strength of sustainable concrete incorporating palm oil fuel ash

Abstract Accurate prediction of compressive strength is important for the efficient design of sustainable concrete incorporating palm oil fuel ash (POFA). This study developed a hybrid artificial neural network model optimized using biogeography based optimization (ANN-BBO) to predict the compressive strength of POFA-based concrete using a compiled literature database of 469 mixtures collected from 22 independent studies. The proposed framework was intended to improve predictive performance relative to a conventional standalone ANN by combining nonlinear learning capability with metaheuristic parameter optimization. The models were developed using six mixture design variables: cement content, POFA content, superplasticizer dosage, coarse-to-fine aggregate ratio, water-to-binder ratio, and curing age. Model performance was evaluated using multiple statistical measures, including the coefficient of determination (R2) and root mean square error (RMSE). For model development, the dataset was divided into training, validation, and testing sets using a 70/15/15 ratio. In addition, model performance was further examined using tenfold cross-validation and bootstrap resampling. Among the evaluated models, ANN-BBO showed the best overall predictive performance, achieving an R2 of 0.983 and a mean absolute error of 2.28 MPa, while also performing favorably compared with the standalone ANN and related models reported in previous studies. The cross-validation results showed that ANN-BBO achieved a mean R2 of 0.954 and a mean RMSE of 4.658 MPa, compared with 0.907 and 6.728 MPa for the ANN model. In addition, more than 60% of the ANN-BBO prediction errors fell within the range of [− 5%, 5%], whereas the corresponding proportion for the ANN model was about 39%, indicating more consistent predictive performance of the hybrid model. Interpretability analysis further indicated that curing age and water-to-binder ratio were the most influential variables governing compressive strength prediction. Overall, the results suggest that the ANN-BBO model may provide a useful tool for rapid strength estimation and preliminary evaluation of concrete containing POFA mixtures within the scope of the compiled dataset.