Prediction of California Bearing Ratio (CBR) of Soils Using AI-Based Techniques

The California bearing ratio (CBR) is an important input parameter in the design of flexible pavements. CBR is often determined in the laboratory involving a laborious and time-consuming testing procedure. In recent years, artificial intelligence (AI) and machine learning (ML) techniques have gained popularity in geotechnical engineering and can circumvent the laborious process of conducting laboratory testing to determine soil properties. This study presents the application of two AI models, viz., random forest regressor (RFR) and artificial neural network (ANN), to determine CBR based on soil basic and mechanical properties such as gradation, maximum dry density (MDD), optimum moisture content (OMC), liquid limit (LL), and plastic limit (PL). A large dataset of 652 data points was gathered from an extensive literature review consisting of all the basic and mechanical properties of soil along with the CBR value. The findings from the study reveal that the RFR model gave a high prediction performance with the coefficient of determination (R²) and mean squared error (MSE) equal to 0.92 and 16.2 respectively, whereas the ANN model resulted in the coefficient of correlation (R) and MSE equal to 0.95 and 28, respectively. Furthermore, sensitivity analysis was carried out to evaluate the most influencing soil parameters affecting the CBR. The results show that MDD has the greatest influence, followed by the percentage of fines, whereas PL has the least importance.