Estimation of permeability of soil using easy measured soil parameters: assessing the artificial intelligence-based models

ABSTRACT In this study, the support vector machine (SVM) technique was used to estimate the permeability of the soil. The performance of SVM model was equated with another type of artificial intelligence techniques such as Gaussian process (GP), random forest (RF) and multi-linear regression (MLR)-based models. While a GP, RF, and MLR model gives a good estimation performance, the SVM model outperforms them. Two kernel functions, Pearson VII and radial basis kernel function, were used in SVM and GP regression models. For this study, a data set contaning 95 observations collected from laboratory experiments. Out of 95, a total of 66 data sets were selected for preparing different algorithms whereas rest 29 data sets were selected to test the models. Input variables include percentage of sand (S), percentage of fly ash (Fa), specific gravity (G), time (T) and head (H), whereas coefficient of permeability (k) is considered as output. A comparison was also done with the various study, which shows poor relationship between the values of permeability. Sensitivity analysis concludes that the parameter, time and water head are the most effective parameters for the estimation of permeability of soil.