Estimation of a ground motion model for induced events by Fahlman's Cascade Correlation Neural Network

Abstract In this paper, a ground motion model (GMM) estimation by Fahlman's Cascade Correlation Neural Network is proposed. We start with a linearised regression estimator and improve the GMM for the Legnica-Glogow Copper District in Poland by adding levels with nonlinear neurons. This gradually leads to the creation of a deep Artificial Neural Network GMM. The paper focuses on the selection of criteria for stopping the development of the network, including minimising the prediction risk, which is defined as the estimation of the quality of the GMM in future observations. Therefore, the solution presented here only applies to the simplest case, which considers only magnitude and distance.