Abstract: A machine learning-based method is developed based on the liquefaction-induced lateral displacement database of Youd et al., 2002, and applied to the simulation of some soil displacement cases in recent earthquakes. We collect the histories of these cases, and then predict them using the existing engineering experience methods to explore the applicability of the model, showing the Youd 2018 method has a good performance. To obtain an optimal machine learning model, this paper discusses the applicability of five models－BP neural network (BPNN), radial basis neural network (RBF), decision tree (DT), random forest (RF), and support vector machine (SVM). We find that the performance of RF is superior to the other machine learning methods. It has high computational efficiency and good data scalability, and can well reflect the characteristics of the data available to this study. Its prediction accuracy is increased by 18.17% relative to the Youd 2018 method. In addition, a sensitivity analysis is carried out of the influencing factors of liquefaction-induced lateral deformation simulation using the RF model.

Key words: soil liquefaction, lateral displacement, data-driven, machine learning, sensitivity analysis