Predicting Ground Granulated Blast-Furnace Slag (GGBS) Concrete Strength Based On PerformanceWith Interpretable Artificial Neural Networks (ANN)

Accurate prediction of concrete strength is essential for effective mix proportioning, performance optimisation, and sustainable construction. Conventional empirical and regression–based approaches often have limited capability to capture the complex and nonlinear interactions among constituent materials, particularly in blended and geopolymer concrete systems. This study develops an Artificial Neural Network (ANN)–based machine learning framework for predicting the compressive and flexural strengths of GGBS–based concrete. A dataset comprising 150 concrete mixtures with varying binder compositions, material proportions, and curing periods was employed to develop and evaluate the predictive models. The modelling framework was assessed using k–fold cross–validation, statistical error indicators, \ and residual diagnostics to examine prediction accuracy, robustness, and generalisation capability. The ANN model demonstrated excellent predictive performance for compressive strength, achieving high coefficients of determination and low prediction errors across the training, validation, and testing datasets. In comparison, the prediction of flexural strength showed relatively lower generalisation performance, which may be attributed to greater experimental variability and the sensitivity of flexural behaviour to crack initiation, propagation, aggregate–matrix interaction, and material heterogeneity. To improve model interpretability, SHAP (SHapley Additive exPlanations) analysis was employed to quantify the contribution of individual input variables. The results identified GGBS content and the alkali–to–binder ratio as the dominant factors influencing predicted strength, demonstrating consistency with the underlying chemical and mechanical behaviour of GGBS–based concrete. The integrated ANN–SHAP framework provides both accurate prediction and meaningful interpretation of strength development. Overall, the findings demonstrate that ANN based machine learning can serve as a reliable and efficient tool for strength prediction and performance–based optimisation of GGBS–based concrete, potentially reducing the need for extensive experimental trials while saving time, material consumption, and associated costs.