A new neural network-based prediction model for Newmark's sliding displacements

The present work aims at developing a new neural network-based prediction model for Newmark's sliding block displacements. The model is developed to predict slope displacement for given earthquake magnitude, focal mechanism, rupture distance, average shear wave velocity of the top 30 m of soil, and critical acceleration of the slope. The network architecture constitutes three layers (only one hidden layer) with nodes per layer 5-5-1. Thus, the network comprises of 36 unknown coefficients. The prediction model utilizes a total of 13,707 data points. Furthermore, inter- and intra-event residuals are evaluated using a mixed-effects algorithm and found to be unbiased, having respective standard deviation accounting to 0.837 and 1.645. The developed slope displacement prediction model is observed to capture the known displacement features, and the patterns are in agreement with the available relations in literature. The applicability of the new model in the estimation of slope displacements hazard is also demonstrated for a representative site in the Himalayan region.