A backpropagation approach for predicting seismic liquefaction potential in soils

Neural networks are successfully used to model the complex relationship between seismic and soil parameters, and the liquefaction potential. Actual field records were used in the analysis. The performance of the neural network models improves as more input variables are provided. The model consisting of 8 input variables is the most successful. These variables are: the SPT value, the fines content, the mean grain size, the equivalent dynamic shear stress, the total stress, the effective stress, the earthquake magnitude, and the maximum horizontal acceleration at ground surface. Comparisons indicate that the neural network model is more reliable than the method of Seed et al. (1985)