Use of Machine Learning to Predict California Bearing Ratio of Soils
Semachew Molla Kassa, Betelhem Zewdu Wubineh
Abstract
CBR is a crucial metric used to assess the durability of base course materials and subgrade soils in various types of pavements. In this research, the machine learning (ML) approach has been implemented using random forest (RF), decision tree (DT), linear regression (LR), and artificial neural network (ANN) models to estimate CBR (California bearing ratio) values of the soil based on seven predictors such as maximum dry density, soil classification, optimum moisture content, liquid limit, plastic limit, plastic index, and swell, which can be easily determined from the laboratory. AASHTO M 145 was used to categorize 252 soil samples that formed the basis of an experimental data set. In this model study, the data were split into 20% test data and 80% training data. Standard statistical measures including coefficient of determination, correlations, and errors were used to assess the effectiveness of the models such as MSE (mean squared error), MAE (mean absolute error), and RMSE (root mean square error). From these evaluation metrics, the random forest algorithm gets a smaller error and larger relative error (R2) value to compare with other algorithms. Therefore, it can be concluded that a random forest algorithm based on the analysis findings can accurately forecast the soil’s CBR.