A comparative analysis of gradient boosting algorithms for predicting the shear stress of root-reinforced unsaturated soil

Bio-stabilisation with plant roots offers a sustainable solution in geotechnical engineering, yet accurately modelling the mechanical behaviour of these composites remains a challenge. This study evaluates the performance of four gradient boosting algorithms – CatBoost, XGBoost, LightGBM and GradientBoosting – for predicting the stress–strain response of unsaturated clayey silt reinforced with Nandina domestica roots. A comprehensive dataset was generated from direct shear tests conducted under varying moisture content (W), root weight density (RWD), depth and normal stress (NS). Using a 70/30 training-testing split, all four models showed robust generalisation in 5-fold cross-validation (mean R2 = 0.90–0.92). On the held-out test data, the models yielded exceptional accuracy. GradientBoosting was the top-performing model with a coefficient of determination (R2) of 0.998, a root mean square error (RMSE) of 0.64 kPa and a mean absolute error (MAE) of 0.43 kPa. The remaining models also performed strongly, with R2values from 0.996 to 0.997. These results validate the capacity of boosting algorithms to capture the complex, non-linear behaviour of root-soil systems, positioning them as powerful and efficient tools for geotechnical analysis and design.