Prediction of compaction parameters of coarse grained soil using multivariate adaptive regression splines (MARS)

In this paper, different models are developed to estimate the compaction parameters of sandy soil using artificial neural network (ANN), least square support vector machine (LS-SVM), and multivariate adaptive regression splines (MARS). The experimental database of Mujtaba et al. (2013) is used for the analysis. The above techniques have been used to improve the regression results. The model equations are established and compared with the regression equation. The MARS model results found to be more accurate and it improved the coefficient of determination to more acceptable levels of 0.88 and 0.81 for the prediction of compaction parameters maximum dry density (gamma(dmax)) and optimum moisture content (omega(opt)), respectively. The results showed that variation between experimental and predicted values of gamma(dmax) is within +/- 4% and that of the omega(opt) is within +/- 13% at 95% confidence level. Sensitivity analysis is carried out to evaluate the parameters affecting the output.