Using Gradient Boosting Regression (GBR) model in predicting the optimum moisture content of soil used for roadbed compaction

Abstract

Optimum Water Content of soil (OWC) is a critical engineering parameter that directly affects compaction efficiency and the long-term stability of subgrade soil in road construction. Determination of OWC using conventional laboratory testing methods is time-consuming and costly, which poses limitations in practical construction. This study applies the Gradient Boosting Regressor (GBR), an advanced machine learning algorithm, to predict OWC based on the physical and mechanical properties of soil samples. Quantitative performance metrics, including the correlation coefficient (R), mean absolute error (MAE), and root mean square error (RMSE), are employed to evaluate model performance. The results show that the model achieves R of 0.916 for the training dataset and 0.776 for the testing dataset, along with low MAE and RMSE values, indicating high predictive accuracy and strong generalization capability. Model interpretability analysis using SHAP techniques and Partial Dependence Plots (PDPs) highlightsthe significant influence of variables such as plastic limit and gravel content on OWC. This research provides an effective supporting tool for subgrade design and construction, while also expanding the application of machine learning approaches in the field of geotechnical engineering.