Comparative study of machine learning algorithms for predicting the compressive strength of geopolymer concrete

Abstract Development of geopolymer concrete (GPC) is vital for minimizing the environmental impact of conventional cement. This study evaluates the predictive capabilities of three machine learning models, response surface methodology (RSM), artificial neural network (ANN), and support vector machine (SVM) for estimating the 28‐day compressive strength of GPC. A total of 22 mix combinations were prepared in two grades (G40 and G60) using fly ash and ground granulated blast‐furnace slag as primary binders. To improve performance, fly ash was partially replaced with rice husk ash and silica fume at 5%, 10%, and 15%. Specimens were ambient‐cured, and compressive strength was measured at 7, 14, and 28 days. Modeling focused on 28‐day results. RSM showed the highest correlation ( R 2 = 0.9805), followed by SVM (0.9656) and ANN (0.9444). However, SVM outperformed others in error metrics, achieving over 58% lower prediction error than ANN and over 10% better performance than RSM. SVM was further optimized using a fine Gaussian kernel and fivefold cross‐validation. While RSM showed the best fit, SVM demonstrated superior generalization and accuracy, making it the most reliable model for predicting GPC compressive strength.