Neural network-assisted generic predictive models of safety factor and yield acceleration for seismic slope stability and displacement assessments

As two well-recognized approaches for seismic slope stability assessment, the pseudo-static analysis estimates the factor of safety (FS) and the Newmark-type analysis estimates permanent downslope-displacement based on input yield acceleration (ky). However, FS and ky are usually obtained from non-trivial slope stability calculations, which can become computationally demanding in probabilistic analyses or regional landslide mapping. This study presents neural network-assisted predictive models for (1) seismic or static FS and the category of failure mode; and (2) ky and the thickness of failure mass. Extensive stability analyses of more than 741,000 and 123,000 slope configurations are conducted to compile datasets of FS and ky, respectively. Performance evaluations indicate that the models produce physically reasonable prediction trends and have good generalization capability with correlation coefficient higher than 0.94 in blind tests. Compared to the existing infinite slope model and predictive tools, the new models achieve improved applicability and functionality, accounting for pore-water pressure, depth to hard stratum, and various failure modes. Both the spreadsheet and MATLAB files established in this study are provided to facilitate generic applications. Therefore, this work not only demonstrates the neural network capability, but also provides useful tools for practitioners, contributing to both the pseudo-static and Newmark-type approaches.