Performance evaluation and machine learning modeling of geopolymer fly ash masonry bricks

Abstract

The present study focuses on the development of sustainable unfired masonry bricks stabilized through geopolymerization of Class C and Class F fly ash using alkaline activators. The experimental program investigates the influence of alkali type (NaOH and KOH), molarity (6 M, 8 M, and 10 M), and curing temperature (ambient to 60 °C) on the mechanical and durability properties of geopolymer bricks. Engineering properties including dry and wet compressive strength, water absorption, efflorescence, and acid resistance were evaluated in accordance with relevant Indian Standards. The developed bricks achieved maximum dry compressive strengths of 21.83 MPa and 17.1 MPa for Class C and Class F fly ash respectively, with water absorption values well within codal limits. Acid resistance tests indicated less than 1% mass loss even at pH 2, demonstrating excellent durability. In addition to experimental evaluation, a machine learning framework was implemented to predict compressive strength using experimentally derived datasets comprising 210 samples. Input parameters included fly ash class, alkali type, molarity, activator ratio, curing temperature, and curing duration. Three supervised regression models—Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Network (ANN)—were developed and validated using fivefold cross-validation. Performance evaluation based on R², RMSE, MAE, and MAPE demonstrated that the Random Forest model provided the highest predictive accuracy, followed by SVR and ANN. Feature importance analysis revealed that molarity and curing temperature were the most influential parameters governing compressive strength development. The integration of experimental investigation with machine learning prediction offers a reliable and efficient approach for optimizing geopolymer brick mix design and enhancing sustainable construction practices.