Data-driven study of nano-silica effects on soil strength using causal models

Nano-silica (NS) has emerged as a promising sustainable stabiliser for improving the mechanical performance of fine-grained soils; however, existing studies predominantly rely on empirical correlations without causal understanding or uncertainty quantification. This study investigates the unconfined compressive strength enhancement of CL–ML soil stabilised with 0%–4% NS under curing periods of 7–84 days, temperatures of 20°C–50°C, pH of 7.58–8.52, and electrical conductivity of 746–1120 µS/cm. A hybrid framework integrating Bayesian causal modelling, symbolic regression, and 3D surface analysis is proposed. SHAP analyses identify curing duration and NS dosage as dominant factors, while Bayesian inference quantifies uncertainty and causal influence. The proposed methodology advances beyond black-box prediction by offering mechanistic interpretability and decision-oriented insights, thereby supporting sustainable and low-carbon geotechnical infrastructure development in line with SDG 9: Industry, Innovation and Infrastructure.