California Bearing Ratio (CBR) is used to assess bearing capacity, deformation characteristics of roadbed soil, and base layer material in pavement structure. In general, CBR is often determined by laboratory or in-situ tests. However, it is time- and cost-consuming to conduct this experiment because this test requires cumbersome equipment such as a compressor. In this study, two Artificial Intelligence models are developed, including a simple model (Decision Tree Regression, DT) and a hybrid model (AdaBoost - Decision Tree, AB-DT). Using 214 data samples from Van Don - Mong Cai expressway, Vietnam, 10 input variables of the model were considered namely particle composition (content of gravel (X1), coarse sand (X2), fine sand (X3), silt clay (X4), organic (X5)), Atterberg limits (Liquid limit (X6), Plastic limit (X7), Plastic index (X8)), and compaction curve (optimum water content (X9) and maximum dry density (X10)). The developed models were evaluated by using a variety of statistical indicators, including coefficient of determination (R2), Root mean square error (RMSE), and Mean absolute error (MAE). The results show that AB-DT model has higher accuracy than the DT model. Moreover, the SHAP value analysis shows that the variable X10 influences the CBR value the most. Thus, it implies that applying the AB-DT model to effectively predict the CBR of the roadbed soil saves time and money for experiments.
Read full abstract