Seismic Slope Displacements: Insights from Traditional Regression and Artificial Neural Networks

Generic predictive models of earthquake-induced slope displacement are developed through classical regression analysis and artificial neural networks (ANNs). The maximum displacement on the slope surface at the end of shaking was computed by finite element simulations of 49 slope models subjected to 1,051 earthquake motions. Predictive models of seismic displacement are developed that characterizes the slope in terms of its yield acceleration (k_y), the natural period of slope (T_slope), and the relative thickness of the slip surface to the height of the slope (H_ratio), and characterizes ground shaking in terms of different ground-motion intensity measures. Across five intensity measures and 10 combinations of intensity measures, peak ground velocity (PGV) is found to be the most efficient and proficient parameter for the displacement prediction, leading to significantly small aleatory variability that is similar to the values derived from the use of multiple intensity measures. The models derived from ANN are similar to those developed from classical regression, although with slightly smaller variability and they did not require development of a complex functional form. These results indicate that ANN may be a viable alternative to classical regression for seismic slope stability models.