Sustainable geopolymer concrete evaluation using machine learning and PDP analysis: Comparative insights into strength prediction

The imperative to mitigate the carbon emissions associated with cement production has accelerated the development of geopolymer concrete (GPC) as a sustainable alternative to Ordinary Portland Cement (OPC). This study benchmarks widely used machine learning models (DT, RF, ANN, and LR) for predicting the compressive strength of geopolymer concrete (GPC) and integrates Partial Dependence Plot (PDP) analysis to connect predictions with binder chemistry and curing dynamics. A dataset comprising 400 experimental samples was compiled, incorporating six critical variables: coarse aggregate, fine aggregate, ground granulated blast-furnace slag (GGBS), sodium silicate, superplasticizer, and curing temperature. Four ML algorithms Decision Tree (DT), Random Forest (RF), Artificial Neural Network (ANN), and Linear Regression (LR) were systematically evaluated. The models were trained and tested on 70/30 ratio, with performance assessed using the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The results indicated that DT outperformed all models, achieving R 2 = 0.97 (training) and 0.94 (testing) with minimal errors (RMSE = 1.85–2.45 MPa, MAE = 1.20–1.70 MPa, MAPE = 0.058–0.118). RF also demonstrated high reliability, R 2 = 0.94, whereas ANN exhibited moderate predictive capacity (R 2 = 0.91-0.89) but was prone to overfitting. LR consistently underperformed (R 2 = 0.88-0.85), reflecting its limitations in capturing the nonlinearities. Partial Dependence Plot (PDP) analysis confirmed that GGBS (Δ = 55 MPa) and curing temperature (Δ = 23 MPa) were the most influential factors, followed by sodium silicate and superplasticizer, whereas excess aggregates reduced CS. This study highlights the practical advantages of tree-based models in AI-driven mix optimization which provides an interpretable and robust framework for sustainable GPC design, thereby promoting the adoption of low-carbon binders in structural applications.