Prediction of lateral spread displacement: data-driven approaches

Site seismic hazard (SSH) is an integral component of seismic risk assessment of engineered structures. The SSH encompasses the effect of ground shaking, landslide, and liquefaction. Discernment of liquefaction and lateral spreading vulnerability is a complex and nonlinear procedure that is influenced by model and parameter uncertainty. In this study, nine different data-driven models were investigated to predict the lateral spread displacement over a free-face and ground-slope conditions. These models include: multivariate adaptive regression splines, generalized additive model, neural networks, generalized linear model, robust regression, regression tree, support vector machine, projection pursuit, and random forest. The results demonstrate efficacy of the proposed models for lateral spreading estimation and in general, the random forest showed a better prediction. Sensitivity analysis is also performed to identify parameters that contribute to the model variability.