Evaluation of Load-Bearing Capacity of Lateritic Soils under Unsoaked and Soaked Conditions Commonly Employed in Pavement Design and Construction in Akure, Ondo State, Nigeria

The objective of this paper was to evaluate the Load-Bearing Capacity of Lateritic Soils under Unsoaked and Soaked Conditions Commonly Employed in Pavement Design and Construction in Akure, Ondo State, Nigeria, using CatBoost and XGBoost . Results obtained show that for the Unsoaked CBR, the CatBoost model achieves an R-squared of 0.78, while XGBoost outperforms it with an R-squared of 0.94. For the Soaked CBR, CatBoost reaches an R-squared of 0.67, and XGBoost achieves 0.77. These results demonstrate the models' strong ability to explain the variance in CBR values. Feature importance analysis reveals that, for Unsoaked CBR, particle size distribution, plasticity limits, and Maximum Dry Density (MDD) are key predictors. Under Soaked conditions, both models highlight the Plasticity Index, Liquid Limit, and MDD, with XGBoost placing greater emphasis on the Plasticity Index. The RMSE values indicate high model accuracy, with the CatBoost Unsoaked model yielding an RMSE of 3.40 and XGBoost achieving 1.82. For Soaked CBR, the RMSE values are 5.03 (CatBoost) and 4.24 (XGBoost), with XGBoost showing greater prediction accuracy. Learning curves demonstrate the models' stable predictive performance with increasing data size. In conclusion, XGBoost is identified as the more effective model for CBR prediction, offering valuable insights for geotechnical engineering, particularly in soil classification and pavement design.