Analyzing the Behavior of Geopolymer Concrete with Different Novel Machine-Learning Algorithms

Abstract

The extensive use of cement in construction industries drastically increases the carbon footprint, but geopolymers as a sustainable alternative can mitigate the environmental impact. This study completely replaces cement with fly ash as the binding material and evaluates how numerous input parameters influence the compressive strength (CS) of geopolymer concrete (GPC). Moreover, the study addresses and employs machine learning models to predict output parameters accurately. The study uses extreme gradient boosting (XGB) and extra tree regressor as a base model and metaheuristics algorithms like water strider optimization and Aquila optimizer (AO) for the optimization of hyperparameters and predict the CS of GPC predicted on 14 input parameters. Furthermore, 10-fold cross-validation reduces overfitting issues for all modal combinations, and AO-XGB demonstrates the best result during training and testing sets, with R2 values 1 and 0.943, respectively. Additionally, the proposed hybridized model is subjected to the model explainability using shapely additive explanations, individual conditional expectation, partial dependence plot, and local interpretable model-agnostic explanations plots, confirming that the model can capture the insights of the feature. Further, the model performance is validated using additional experimental data with over 90% accuracy. These insights make understanding the connections between the various elements included in thorough evaluations of geopolymers for implementing sustainable practices easier and lay the foundation for future studies relating to GPC and machine learning.