hide
Free keywords:
-
Abstract:
We use a number of machine learning models to predict the seismic site amplifications of shallow bedrock sites calculated from one-dimensional ground response analyses. We compare the results with a regression-based model that are conditioned on VS30, site period, and peak ground acceleration. We show that use of the array type of information for the ground motion and site profile in machine learning training produces excellent predictions of the surface response. In contrast, employing parameters typically used for the site amplification models widely employed in ground motion prediction equations produces only marginal improvement compared with a regression-based model. The fit can be improved through utilization of additional parameters, although the improvement is not as dramatic as when using the array data.
