Sustainable approach of strength measurement for soil’s stabilized with geo-polymer with hybrid ensemble models

Abstract

This study introduces an innovative method for soil stabilization by integrating geopolymer binders with a non-iterative hybrid ensemble modeling framework. Using 270 samples, key parameters such as molarity, silt content, and chemical ratios (Si/Al, Na/Al) were analyzed to predict the unconfined compressive strength (UCS) of geopolymer-stabilized clay. Five machine learning models Random Forest, Support Vector Regression, Extreme Learning Machine, Artificial Neural Networks, and Multivariate Adaptive Regression Splines were developed and combined in a unique hybrid ensemble. This approach eliminates the need for iterative optimization, ensuring computational efficiency. Among the hybrid models, Hybrid Random Forest (HRF) and Hybrid Artificial Neural Network (HANN) delivered superior performance, with HRF achieving an R² of 0.997 (training) and 0.983 (testing), and an RMSE of 0.784 MPa. Additionally, HRF exhibited an Index of Agreement (IOA) of 0.996, indicating exceptional alignment with actual UCS values. A20-index scores for HRF and HANN were 0.982 and 0.973, respectively, highlighting their robustness across diverse conditions. The study’s non-iterative hybrid ensemble approach offers flexible and accurate UCS predictions while reducing computational demands. This research emphasizes the potential of geopolymer materials and advanced machine learning models to advance sustainable and efficient soil stabilization solutions in geotechnical engineering.

Keywords

• Sustainable unconfined compression strength
• Machine learning geo-polymer clay