Abstract
This study investigates the application of soft computing techniques for predicting the Rapid Chloride Penetration Test values in geopolymer concrete incorporating supplementary cementitious materials such as silica fume, fly ash, ground granulated blast furnace slag, and microfibers. Four machine learning models- Adaptive Boosting (AdaBoost), African Vultures Optimization Algorithm (AVOA), Categorical gradient Boosting (CatBoost), and LightGBM Regressor (LGBMR) were employed to analyze the chloride permeability behaviour. The results demonstrated high predictive accuracy, with CatBoost achieving the best performance (training R(2) = 0.9982, testing R(2) = 0.9640), followed by AVOA (R(2) = 0.9854 training, 0.9400 testing), LGBMR (R(2) = 0.9894 training, 0.9618 testing), and AdaBoost (R(2) = 0.9282 training, 0.8922 testing). SHAP analysis revealed the relative influence of each material on chloride resistance, with GGBS and silica fume showing significant contributions. The findings highlight the potential of soft computing techniques in optimizing durable and sustainable geopolymer concrete, reducing reliance on experimental trials.