Sustainable stabilization of marine soft clay using low-carbon binder with high-volume industrial waste: Performance, microstructure, and explainable ML prediction

This study aims to reduce the carbon footprint and energy consumption of ordinary Portland cement (OPC) by developing a low-carbon cementitious material (LCM) for the stabilization of marine soft soil. The LCM consists of 20% OPC and 80% industrial waste, predominantly ground blast furnace slag (GB). The effects of calcium carbide residue (CR), rice husk ash (RA), and phosphogypsum (PG) on the strength and durability of LCM-stabilized marine soft soil were systematically investigated. Hydration products and microstructural evolution were analyzed through pH measurements, X-ray diffraction (XRD), scanning electron microscopy (SEM), and mercury intrusion porosimetry (MIP). Additionally, the strength prediction was carried out by explainable machine learning, and the dependence of chemical composition on the strength target variable was obtained. The results indicate that the inclusion of CR elevated the pH of the stabilized soil, while RA had a minimal effect on strength evolution. In contrast, PG significantly enhanced the strength of the stabilized soil. When compared to pure OPC stabilization at an identical total binder content of 15% soil mass and size of 50 mm×50 mm cylinder under standard curing, the highest compressive strengths of the LCM-stabilized soil cured for 7, 28, and 60 days reached 89.7%, 112.5%, and 158.9% of the OPC-based strengths, respectively. Moreover, under simplified assumptions regarding material unit prices and CO 2 emission factors (excluding transportation and pretreatment impacts, and assigning zero emissions to some by-products), a notional LCM mix was estimated to achieve cost and CO₂ emission reductions of up to 44.4% and 82.5%, respectively, compared to OPC-based stabilization. These figures should be interpreted as theoretical upper bounds rather than guaranteed field values, as they are sensitive to project-specific supply chains and material sourcing. Microstructural analysis revealed that increasing PG content promoted the formation of AFt and C-S-H crystals, which enhanced the compactness of the microstructure, reduced pore size, and consequently improved strength. Explainable machine learning, using Shapley Additive explanations, identified key factors influencing strength development, including SO 3 content, age, and CaO. A comparative evaluation of ensemble models indicated that the Gradient Boosting Regressor (GBR) and XGBoost showed marginally better predictive performance than Random Forest (RF) and LightGBM for this dataset. These findings from microstructural analysis and explainable machine learning provide a deeper understanding of the low-carbon cementitious material's stabilization mechanism. • A low-carbon cementitious material (LCM) composed of 20% OPC and 80% industrial waste was applied for stabilization of marine soft soil. • Phosphogypsum significantly enhances the initial strength gain but decreases the immersion and wetting-drying strength of LCM stabilized marine soft soil. • The filling and cementation effects on microstructure evolution were revealed. • GBR and XGBoost models show marginally better strength prediction performance.