Application of multi-expression programming (MEP) in predicting the soaked California bearing ratio (CBR) value of fine-grained soil

This study is attempted to develop the prediction model for the soaked California bearing ratio (CBR) value of fine-grained soil through the multi-expression programming (MEP) approach. The modeling phase was performed on 1011 in situ collected soil samples from an ongoing highway construction project work site. Based on the literature recommendations and present study database analysis, six relevant input parameters were extracted from numerous geotechnical parameters achieved through laboratory experiments. Using those parameters, several tentative combinations were prepared to develop the most efficient predictive model. The comparative results of all the models demonstrate the higher accuracy with PL, PI, S, FC, MDD and OMC as input parameters. The developed model with the adopted input parameters is able to explain 63% variability in the soaked CBR value of fine-grained soil. Almost 97% of total observations were found to be predicted within +/- 20% variations. Additionally, the sensitivity analysis reveals that the soaked CBR value is prominently influenced by the MDD followed by FC, PL, S, PI and OMC. The validation results of the model exhibit that the developed model is also worthy in predicting the soaked CBR value of the unseen dataset. Eventually, a graphical user interface was generated for the future convenience of the researchers and site engineers.