Modelling Soil Water Retention Curve for Cohesive Soil Using Artificial Neural Network

The determination of soil water characteristics or retention curves (SWCC) from laboratory procedures is tedious and time-consuming. In this study, the prediction of Fredlund and Xing fitting parameters applicable to finding the suction potential of plastic soils is determined using an artificial neural network modelling. A three-layer ANN model consisting of input, hidden, and output layers. The input layer consists of the highly influencing soil variables such as percentage of soil passing 4.75 mm and 0.075 mm IS sieve, liquid limit, plasticity index, and saturated volumetric water content. The hidden layer consists of 20 neurons to train the input data with the existing output data using the Levenberg–Marquardt backpropagation method. The Fredlund–Xing (FX) model fit parameters, namely af, bf, and cf, were selected as the output variables, and the ANN model with 70% dataset for training, 15% to validate, and 15% to test the network gives a higher correlation of 0.65–0.82. Thus, the use of the ANN model confirms the ability of selected input variables to predict the fitting parameters of the FX model. The study confirms the ability of ANN as an aiding tool to determine the suction potential of plastic soils such that expensive and cumbersome laboratory testing procedures can be replaced.