Abstract
In this research, a reliable Cone Penetration Test data set with a wide range of parameters was integrated in an Artificial Neural Networks (ANN) program in order to evaluate the liquefaction potential of soils. This research proposed three ANN models with different input parameters and one output parameter which represent the occurrence and non- occurrence of liquefaction. The results of this research showed that the complex relationship between the soil, stress and earthquake parameters was well understood by the proposed ANN models. Moreover, the success rate in prediction liquefaction was higher than that given by the traditional methods. The research results showed that preprocessing, normalizing or calibrating the data before assessing the liquefaction potential is not needed as requested by previous works. In view of the relative importance of effective parameters in liquefaction assessment, it is found that q(c) has a more important role than sigma(vo) and sigma'(vo).