A machine learning framework for predicting the unconfined compression index of nano-silica stabilized fine-grained soils using leave-one-soil-type-out validation

The Lesser Himalayan region presents significant challenges for infrastructure development due to weak, landslide-prone soils. This study evaluates nano-silica (NS) stabilization of three fine-grained soils and introduces the Unconfined Compression Index ( Ic ), defined as the axial strain at 50% of peak unconfined compressive strength ( qu ), as a measure of stabilized-soil deformability. Laboratory specimens were prepared with NS contents of 0–4% and cured for 3–90 days. Six machine-learning models Linear Regression, Random Forest, Extreme Gradient Boosting (XGB), K-Nearest Neighbors, Support Vector Machine, and Lasso Regression—were trained to predict Ic from soil type, NS content, curing duration, and early-cure qu . To avoid interpolation bias inherent in random splitting of factorial data, a Leave-One-Soil-Type-Out Cross-Validation was adopted. XGB achieved the highest predictive accuracy ( R 2 = 0.961; MAE = 0.158), while multicollinearity among predictors was mitigated via regularization. The results demonstrate that Ic can be estimated from routine strength data, offering a practical tool for assessing strain compatibility in NS-stabilized Himalayan soils.