Deep neural network based approach for site amplification prediction of shallow bedrock sites

Site amplification models have been proposed based on the empirical and simulation results and represented by the functional form with regression analysis. However, these regression models result in large differences between the results. In this study, a new site amplification model is developed by using a deep neural network (DNN) to reduce the uncertainties in site amplification predictions. The results of the linear and nonlinear site response analyses are used to train and test the model. The results of the DNN-based model are compared with those of the regression model. The DNN-based model outperforms the regression model and shows lower residuals throughout spectral periods. The DNN model can replace the regression models and reduce the uncertainties caused by simple prediction models.