A generic framework for mix design of geopolymer for soil stabilization: Composition-informed machine learning model

Abstract

Geopolymer has emerged as an environmentally sustainable alternative to cement for soil stabilization. Although machine learning technology exhibits great potential in designing the geopolymer, its current applications are limited to specific types of geopolymer. This study introduces a composition-based method to develop a machine learning model that facilitates the mix design of diverse types of geopolymer. A unique dataset comprising 990 mix designs of geopolymer for soil stabilization was established. Based on this dataset, a composition-informed machine learning model was developed to predict the unconfined compressive strength of geopolymer-stabilized soils. By using the developed model, a generic framework for the mix design of diverse types of geopolymer is proposed. The performance of the developed model and the effectiveness of the mix designs derived from the proposed framework was evaluated by using datasets that are independent of the dataset established in this study. The results show that the developed model enables the reasonable predictions of the strength of geopolymer-stabilized soils. The mix designs formulated based on the proposed framework is comparable to those derived from experimental studies. The proposed framework can serve as a cost-effective and efficient toolkit for the mix design of diverse of types of geopolymer.