Equation-driven strength prediction of GGBS concrete: a symbolic machine learning approach for sustainable development

The growing demand for low-carbon construction materials highlights the need to predict and optimise the performance of Ground Granulated Blast-Furnace Slag (GGBS) concrete with high accuracy. This study aims to develop reliable and interpretable predictive models for estimating the compressive strength (CS) of GGBS-based concrete to support sustainable mix design. Three machine learning approaches, Multilayer Perceptron (MLP), Adaptive Boosting (AdaBoost), and Gene Expression Programming (GEP) were trained on a diverse dataset collected from published experimental studies. Each model was evaluated using five-fold cross-validation and multiple statistical indicators (R², RMSE, and MAE), and interpretability was examined through LIME and permutation-based sensitivity analysis. The MLP model achieved the highest predictive accuracy (R² = 0.89), followed by AdaBoost (R² = 0.88) and GEP (R² = 0.86). The GEP model provided a transparent mathematical equation suitable for practical strength estimation. These findings reduce the need for repetitive laboratory testing and enable data-driven optimisation of GGBS concrete design. The study uniquely integrates explainable artificial intelligence with symbolic regression to combine predictive accuracy and interpretability, offering a reproducible framework for sustainable concrete design and analysis.