Multiscale Optimization of Clayey Soil Stabilization Using Wood Ash and Sodium Chloride Through Experimental Statistical and Machine Learning Approaches

Abstract A hybrid multiscale framework combining synthetic data modeling, laboratory experimentation, and feature-driven machine learning (ML) evaluation was deployed to optimize the stabilization of clayey soils using dual binders—wood ash (WA) and sodium chloride (SC). WA and SC were applied in varying proportions (5–20% by dry weight), and the soil samples were subjected to standard geotechnical testing, including unconfined compressive strength (UCS) and California Bearing Ratio (CBR) assessments after curing periods of 7, 14, and 28 days. UCS values increased significantly with curing time, with a peak strength of ~ 220 kPa at 15% WA and 10% SC after 28 days, compared to ~ 90 kPa in untreated soil—representing over 140% improvement. CBR values also improved from 6% (untreated) to 21% under the optimal mix. Gradient Boosting outperformed other ML models with R² = 0.85, MAE = 2.95 kPa, and RMSE = 3.77 kPa. Feature importance analysis ranked curing time and WA as the dominant predictors. These findings suggest the viability of data-driven and resource-efficient soil stabilization methods for applications in infrastructure development, particularly in resource-constrained or moisture-sensitive regions.