Leveraging ML for predicting UCS of soil stabilised with one-part alkali-activated binder

The transition to sustainable construction materials has driven interest in alternatives to Portland cement. Soil stabilisation with alkali-activated binders is a promising approach, yet its widespread application requires reliable predictive tools for assessing unconfined compressive strength (UCS). This study explores the use of machine learning algorithms to predict UCS in soil stabilised with a one-part alkali-activated binder. An experimental data set was compiled to train and validate multiple machine learning models, including random forests, artificial neural networks, and support vector machines. Despite the data set’s limited size, the models demonstrated strong predictive accuracy, with random forest achieving an R2 exceeding 0.80. Sensitivity analysis revealed that water and soil content were the most influential parameters, aligning with established geotechnical principles. These findings highlight the potential of machine learning as a reliable tool for optimising soil stabilisation techniques. By enhancing predictive capabilities, this approach supports more efficient material selection, reducing reliance on extensive laboratory testing. The study underscores the value of integrating data-driven methods into geotechnical engineering to advance sustainable and high-performance soil treatment solutions.