Machine Learning–Assisted Design of Geopolymer-Stabilized Soils: Predictive Modeling, Key Influencing Factors, and Future Research Directions

Machine Learning–Assisted Design of Geopolymer-Stabilized Soils: Predictive Modeling, Key Influencing Factors, and Future Research Directions

Malihe Esmati1 Mehdi Zabihi Saani Shahr Kohne2 ArminHatamiRad3

1) 1. Master s student in Civil Engineering, Semnan University, Semnan, Iran,
2) 2. Master of Geotechnics, Islamic Azad University of Mashhad,
3) Master’s Student in Civil Engineering, Islamic Azad University, Mashhad Branch, Mashhad, Iran

IOI Code : ABEB-ABHZFA
Publication : 4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
    The stabilization of problematic soils is a fundamental practice in geotechnical engineering aimed at improving the mechanical performance and long-term durability of weak ground materials. In recent years, geopolymer binders produced through the alkali activation of aluminosilicate precursors such as fly ash, ground granulated blast furnace slag, and metakaolin have emerged as sustainable alternatives to conventional stabilizers like Portland cement and lime. These materials offer significant environmental advantages, including reduced carbon emissions and the beneficial reuse of industrial by-products. However, the engineering behavior of geopolymer-stabilized soils is influenced by numerous interacting parameters, including precursor composition, activator concentration, curing conditions, binder dosage, and soil mineralogy. The complex and nonlinear relationships among these variables have motivated the application of machine learning (ML) techniques to predict and optimize the performance of stabilized soils.This paper presents a comprehensive review of recent advances in machine learning–assisted design of geopolymer-stabilized soils. Various predictive models, including artificial neural networks (ANN), support vector machines (SVM), random forest (RF), gradient boosting algorithms, and hybrid data-driven approaches, are critically examined with respect to their ability to estimate key geotechnical properties such as unconfined compressive strength, California bearing ratio, and durability indicators. The influence of input variables and feature importance analyses reported in the literature is also discussed. Furthermore, the study evaluates model performance, data requirements, and current limitations associated with ML-based prediction frameworks. Finally, future research directions are proposed to promote the integration of machine learning with sustainable ground improvement technologies and to enhance the reliability of predictive tools in geotechnical engineering practice.

Keywords : geopolymer stabilized soil machine learning in geotechnical engineering predictive modeling alkali activated binders sustainable soil stabilization