Ground motion amplification models for Japan using machine learning techniques

Abstract

Earthquake-induced ground motions can be altered by various factors that are associated with the characteristics of earthquake sources, paths, and sites. Conventionally, regression approaches have been used to develop empirical prediction models for ground motion amplifications. We developed models for ground motion amplifications based on three machine learning techniques (i.e., random forest, gradient boosting, and artificial neural network) using the database of the records at the KiK-net stations in Japan. The proposed machine learning based models outperforms the regression based model. The random forest based model provides the best estimation of amplification factors. Average shear wave velocity and the depth of the borehole are the two factors that influence the amplification model the most. Maps of the amplification factors for all KiK-net stations under moderate and large earthquake scenarios are provided. The three machine learning technique based models are also provided for the forward prediction of other earthquake scenarios.