Accurate prediction of peak shear strength can facilitate the mechanical assessment of stabilized and reinforced soils while reducing reliance on extensive laboratory testing. This study develops a data-driven framework for predicting and interpreting the peak shear strength of fly ash-amended, lignin-fiber-reinforced silty sand. Eight machine-learning algorithms were comparatively evaluated using 90 observations obtained from unconsolidated–undrained triaxial tests, with lignin fiber content, curing time, initial moisture content, and confining pressure as input variables. Among the evaluated models, Gradient Boosting exhibited the best generalization performance, achieving a test R² of 0.871, an MAE of 92.60 kPa, and an RMSE of 110.10 kPa. Model-based feature importance identified lignin fiber content as the dominant predictor, followed by confining pressure, initial moisture content, and curing time. Partial-dependence and sensitivity analyses further revealed pronounced nonlinear effects of lignin fiber content, whereas confining pressure showed a consistently positive influence on the predicted strength. Residual diagnostics indicated a standard deviation of 113.14 kPa, with no residuals exceeding the adopted ±2σ threshold. Overall, the findings demonstrate the potential of nonlinear ensemble learning for accurate and interpretable prediction of peak shear strength within the investigated experimental domain, while providing insight into the relative influence of key material and testing parameters.
