Application of genetic expression programming and artificial neural network for prediction of CBR

Abstract

The CBR (California bearing ratio) value is an important parameter of the subgrade soil required for the design of pavements. The present study deals with the application of genetic expression programming (GEP) and artificial neural network (ANN) for the prediction of CBR. Various soil properties, such as gravel percentage (G), sand percentage (S), fine content (FC), liquid limit (WL ), plastic limit (WP ), plasticity index (IP ), optimum water content (Wc opt) and maximum dry density (γd max), were considered as the variables input parameters in the analyses. Mathematical expressions were developed for the prediction of CBR and their dependency over different combinations of variables was obtained by GEP. The same combinations of variables were used for ANN prediction. It was observed that both the GEP and ANN methods fit well for CBR prediction and the model consisting of variables G, S, IP , WC opt and γ dmax was found to be the best model. It was found that 80 numbers of chromosomes, 3 head length and 3 numbers of fixed genes is the optimal condition for the prediction of CBR by GEP. The G and S were found to be the most significant parameters with 28.41% and 39.62% contribution in case of GEP and 26.83% and 23.37% in case of ANN, respectively. The variable WP was not used by GEP during optimal model construction, which may imply poor dependability of CBR over it.