Accurate prediction of the unconfined compressive strength (UCS) of geopolymer-stabilized clayey soil is critical for geotechnical engineering. Conventional regression algorithms and even advanced machine learning approaches such as artificial neural networks often struggle to fully capture the highly non-linear interactions among soil properties and geopolymer mix parameters while maintaining computational efficiency and interpretability on limited datasets. To address these challenges, this investigation proposes a novel hybrid predictive framework based on a sector fruit fly optimization algorithm–enhanced extreme learning machine (SFOA-ELM) for improved UCS estimation. The model used 270 experimental records with 8 input variables (i.e., liquid limit, plasticity index, GGBS content, molarity). The baseline ELM achieved moderate performance (training R² = 0.9432, testing R² = 0.9054; RMSE = 1.55 MPa and 1.91 MPa, respectively). The predictive accuracy of the standard FOA-ELM improved to an R² of 0.9768 (training) and 0.9318 (testing), with RMSE values of 0.99 MPa (training) and 1.62 MPa (testing). The performance of the SFOA-ELM model improved to have excellent predictive performance with an R² value of 0.9775 (training) and 0.9446 (testing), and an RMSE value of 0.98 MPa (training) and 1.462 MPa (testing), respectively. The SFOA-ELM model reduced the testing MSE by 42% compared to ELM and by 19% compared to FOA-ELM. A SHAP analysis suggested that GGBS content, plasticity index and liquid limit were the most important for prediction. The statistical validations and residual analysis confirmed SFOA-ELM’s ability to generalize effectively and exhibit a tight distribution of errors. These findings show that the SFOA-ELM framework offers a robust, computationally efficient, and interpretable tool for geopolymer mix design and UCS prediction, providing practical decision-support for geotechnical engineering applications.
Abdellatief et al. (Fri,) studied this question.