Support vector regression-based modeling of cumulative infiltration of sandy soil

Abstract

This paper examines the capability of the support vector machine-based regression approach to predict the cumulative infiltration from sandy soil. A data-set consisting of 413 cumulative infiltration measurements was used in the present analysis. Results drawn from radial basis function (RBF) and polynomial (poly) kernel-based support vector regression (SVR) were compared with multiple linear regression (MLR), M5P tree, generalized regression neural network (GRNN), and two conventional models, Kostiakov model and US-Soil Conservation Service (SCS) model. Out of 413, a total of 289 data were randomly selected for training different algorithms, whereas residual 124 data were used to test the models. The correlation coefficient (C.C) of 0.9837 with root mean square error (RMSE) value of 0.3073 was achieved by RBF kernel-based support vector regression in comparison to C.C value is 0.9255 with RMSE value of 0.6423 through M5P tree model. Comparisons of results propose that RBF based SVR works well. Single-factor ANNOVA results conclude that there is an insignificant difference between observed and predicted values using different models. Sensitivity analyses further suggest that the time is the most important parameter when SVR-based modeling approach is used for prediction of cumulative infiltration.