A Machine Learning Approach to California Bearing Ratio Prediction: Evaluation of GEP and Ridge Regression

One of the most reliable on-site tests in geotechnical science is the CBR test, whose results play an effective role in estimating the bearing capacity of pavements and in designing them more precisely. The goal of this research is to predict the CBR test using the natural characteristics of the soil. To achieve this aim, Ridge Regression and gene expression programming (GEP) methods are employed to develop a new model that is a function of five independent variables: fine-grained, sand percent, gravel percent, optimum moisture content (OMC), and maximum dry density. The correlation coefficient of the best model resulted in 0.90 and 0.65 for GEP, and 0.90 and 0.70 for Ridge for training and validation data, respectively. The obtained results indicate the high accuracy of the presented model, which can be used for predesign purposes.