Slope failure prediction combining limit equilibrium, case histories, and Bayesian Markov Chain monte Carlo Method

This study demonstrates the integration of an analytical geotechnical method and a statistical method to predict the stability of soil slopes using a probabilistic approach. The model utilized Bayesian Markov Chain Monte Carlo re- arametrization, based on prior distributions generated from 104 published case histories, and a synthetic database consisting of 4,032 factor of safety values from limit equilibrium analyses. Validation of the Bayesian model against slope stability case histories showed an area under the receiver operating characteristic curve (AUC-ROC) of 86%, indicating high classification accuracy. The results showed that the Bayesian model performed well when predicting slope stability or instability. It can be used to inform the preliminary design or remediation of slopes by incorporating parameter uncertainties and random effects generally not considered by traditional deterministic studies.