An improvement to MLR model for predicting liquefaction-induced lateral spread using multivariate adaptive regression splines

Soil liquefaction during earthquakes can result in ground movements that cause damage to buildings and lifelines. Lateral spreading is one form of earthquake-induced ground movements that have caused extensive damage in previous earthquakes. The lateral displacement is dependent on many factors including the earthquake magnitude, thickness and particle size of the liquefiable subsoils and the depth of the groundwater. A number of analytical and empirical methods have been proposed to predict the magnitude of the lateral displacement. One common empirical method is the MLR model which is based on multiple linear regression (MLR) analysis of a database of observed case histories. It is proposed in this paper to use a nonparametric regression procedure known as multivariate adaptive regression splines (MARS), as an improvement to the current MLR model to predict the liquefaction induced lateral displacement. First the basis of the MARS method and its associated procedures are explained in detail. Results are then presented to show the accuracy of the proposed approach, in comparison to the commonly used multiple regression approach. Analysis of observed case histories indicated that the MARS outperforms MLR in terms of predictive accuracy. MARS automatically models non-linearities and interactions between variables without making any specific assumptions. Furthermore, it is able to provide the relative importance of the input variables and give insights of where significant changes in the data may occur. (C) 2013 Elsevier B.V. All rights reserved.