The use of neural networks for predicting the factor of safety of soil against liquefaction

In this paper, the Factor of Safety (FS) values of soil against liquefaction was investigated by means of Artificial Neural Network (ANN) and Multiple Regression (MR). To achieve this, two earthquake parameters, namely earthquake magnitude (Mw) and horizontal peak ground acceleration (amax), and six soil properties, namely Standard Penetration Test Number (SPT-N), saturated unit weight ( sat), natural unit weight ( n), Fines Content (FC), the depth of Ground Water Level (GWL), and the depth of the soil (d), varied in the liquefaction analysis; then, the FS value was calculated by the simplified method for each case by using the Excel program developed and utilized in the simulation of the feed-forward ANN model with backpropagation algorithm and the MR model. The FS values predicted by both ANN and MR models were compared with those calculated by the simplified method. In addition, five different performance indices were used to evaluate the predictabilities of the models developed. These performance indices indicated that the ANN models were superior to the MR model in terms of predicting the FS value of the soil.