Enhanced prediction of fly ash concrete compressive strength using machine learning models

Purpose This study aims to improve the prediction of concrete compressive strength (CS) in fly ash-based mixes by applying ensemble and non-ensemble machine learning (ML) techniques. The goal is to identify the most accurate model, reduce reliance on time-consuming laboratory tests and highlight the influence of key mixture parameters on CS to support sustainable concrete design. Design/methodology/approach A dataset of 300 fly ash concrete samples was used, incorporating seven input variables: cement, fly ash, water, superplasticizer, coarse and fine aggregates and curing age. Multiple ML models, including adaptive boosting (AdaBoost), random forest (RF), eXtreme Gradient Boosting (XGBoost), artificial neural networks (ANN), adaptive neuro-fuzzy inference system, gene expression programming, support vector regression, multiple nonlinear regression and multiple linear regression were developed. Their performance was evaluated using R2, root mean squared error (RMSE), mean absolute error (MAE) and k-fold cross-validation, supported by visual diagnostics. Sensitivity analysis was conducted to quantify each input’s contribution to CS. Findings The ensemble models outperformed all non-ensemble approaches. AdaBoost achieved the highest testing accuracy (R2 = 0.960, RMSE = 2.66 MPa), while XGBoost showed the strongest generalization across folds. ANN delivered the best accuracy among non-ensemble models but remained below the ensemble models. Sensitivity analysis revealed curing age (40.59%) and cement content (39.25%) as the dominant predictors of CS. Originality/value This study demonstrates an efficient ML-based framework for predicting CS in fly ash concrete, significantly reducing reliance on experimental testing. The integration of ensemble models and sensitivity analysis provides a robust tool for optimizing sustainable concrete mixes and supports broader adoption of eco-friendly materials in construction.