Application of the LGBM model for predicting the maximum dry density of soil used in roadbed compaction

Abstract

In civil engineering, accurately determining the maximum dry density (MDD) of embankment soil plays a crucial role in ensuring construction quality. This study proposes the application of the Light Gradient Boosting Machine(LGBM) algorithm to predict MDD based on the geotechnical properties of soil. The dataset consists of 214 soil samples collected from the Van Don –Mong Cai expressway construction project (Vietnam) with input parameters including gravel content (G), coarse sand content (CS), fine sand content (FS), silt-clay content (SC), organic content (O), liquid limit (LL), plastic limit (PL), plasticity index (PI), and MDD as the output parameter. Quantitative evaluation metrics such as R², RMSE, and MAE were used to assess the model’s performance. The results indicate that the LGBMachieves good predictive performance with the following metrics on the training dataset: R² = 0.941, RMSE = 0.028, MAE = 0.020; and on the testing dataset: R² = 0.771, RMSE = 0.059, MAE = 0.048. Additionally, model interpretation methods such as SHAP and partial dependence plots (PDP) show that factors like gravel content and coarse sand content contribute to increasing MDD, while other factors tend to decrease it. The findings highlight the significance of adopting a LGBM-based approach for rapid and accurate MDD prediction to support road design and calculation.