Predicting compressive strength of polyurethane-modified light weight geopolymer bricks using hybrid AI models

Abstract

Lightweight geopolymer bricks have emerged as efficient and sustainable architectural materials, offering reduced density, enhanced thermal insulation, and excellent fire resistance. Their lower structural weight improves seismic performance and design flexibility, while their reduced carbon footprint makes them a viable alternative to conventional construction materials. In parallel, Artificial Intelligence-driven predictive approaches have become essential for optimizing material performance and accelerating innovation in sustainable construction. This study investigates the application of advanced hybrid AI models, ANFIS-GA, CNN-LSTM, SVM + K-means, and XGBoost + K-means to predict the compressive strength of polyurethane-modified lightweight geopolymer bricks. A comprehensive dataset of samples was analyzed using statistical and machine learning techniques to evaluate the influence of mix design parameters on compressive strength. The results demonstrate excellent predictive performance, also the Area Under the Curve (AUC) values for all models were remarkably high, ranging between 0.97 and 1.00, for both the training and testing datasets. Among the models, XGBoost + K-means and SVM + K-means achieved superior accuracy, with R ² values up to 0.98 (training) and 0.96 (testing), alongside low error metrics (MAE = 1.00 MPa, RMSE = 2.25 MPa). High prediction reliability was further confirmed through coverage metrics within ±3 MPa and ±5 MPa. These findings highlight the effectiveness of hybrid AI models in predictive performance of polyurethane-modified geopolymer systems and demonstrate their potential for intelligent material design, performance optimization, and sustainable construction applications.