Estimating the permeability coefficient of soil using CART and GMDH approaches

Abstract

Permeability coefficient of soil (k) is one of the most important parameters in groundwater studies. This study, two robust explicit data-driven methods, Including classification and regression trees (CART) and the group method of data handling (GMDH) were developed using the characteristics of soil, i.e., clay content (CC), water content (ω), liquid limit (LL), plastic limit (PL), specific density (γ), void ratio (e) to generate predictive equations for prediction of k. When compared to CART; mean absolute error (MAE) = 0.0051, root mean square error (RMSE) = 0.0088, scatter index (SI) = 64.00%, correlation coefficient (R) = 0.7841, index of agreement (IA) = 0.8830; the GMDH equation produced the lowest error values; MAE = 0.0044, RMSE = 0.0072, SI = 52.17%, R = 0.8493, Ia = 0.9184; in testing stage. Although, GMDH had better performance, however, CART and GMDH could be considered effective approaches for the prediction of k. HIGHLIGHTS Two predictive models were developed to estimate the permeability coefficient of soil.; GMDH and CART algorithms were evaluated in this study.; GMDH provided more accurate results when compared to CART for the prediction of permeability coefficient of soil.; The water content was the most effective parameter for determining the permeability coefficient of soil.; Field data were used in this research.;

Keywords

• decision trees
• explicit models
• infiltration
• k
• soft computing