Numerous studies on predictive modeling methodologies have been conducted in order to evaluate the quality of treated granular materials as a consequence of the increased need for efficient soil stabilization solutions. The purpose of this study is to examine whether or not machine learning ( \(ML\) ) can be used to predict the unconfined compressive strength \(({q}_{u}\) ) of low-quality sand that has been saturated with natural pozzolanic geopolymer. Through the use of the Gradient Boosting Regression ( \(GBR\) ) approach, the link between these factors and \({q}_{u}\) has been successfully identified. Both the Red Fox optimization ( \(RFO\) ) and the Pufferfish optimization ( \(PuO\) ) are examples of complex metaheuristic optimization procedures that are used for hyperparameter tweaking in order to increase the anticipated accuracy and stability of the \(GBR\) model. The findings demonstrate the effectiveness of machine learning-driven approaches in determining the stability of granular materials and provide a valuable analysis for the improvement of geotechnical engineering techniques via the use of data-driven modeling application tools. According on the supplied data, it is likely that both \(GBR\left(Pu\right)\) and \(GBR\left(RF\right)\) will precisely compute \({q}_{u}\) . The effectiveness of the \(GBR\left(Pu\right)\) model exceeds that of the \(GBR\left(RF\right)\) technique about the ultimate objective it aims to achieve.
