Evaluation of rock slopes susceptible to circular failures using logistic and multiple regression models

Abstract

The analysis of stability of slopes is a classical problem for geotechnical engineers. In practice, many methods are available for the desired purpose, from basic kinematic analysis to two- and three-dimensional limit equilibrium analyses and numerical modeling with various user-friendly softwares. However, additional techniques are also required to provide knowledge necessary for decision-making. In this research, a reliable dataset provided by Sah et al. (1994) was used to analyze the stability evaluation of rock slopes subjected to circular failures. For this purpose, using Slide 2018 program, 44 separate limit equilibrium slope models were built for each case given in the original work. The provided material properties and slope geometries in heavily fractured and/or very weak or highly weathered rock masses were considered during the model building stage. It was found that Slide 2018 program generated dissimilar safety factors compared to those given by Sah et al. (1994) for the investigated slope cases. Binary logistic and multiple linear regression techniques were implemented in the study to promote alternative approaches for the prediction of the stability condition and safety factor (SF) of slopes excavated in heavily fractured/highly weathered rock masses. The condition of slopes (stable or failed) was predicted by binary logistic regression model with 90.9% accuracy. The SF of the slopes was estimated by a multiple linear regression model with 95.6% accuracy. It was concluded that both statistical techniques could be sufficiently used as alternative approaches to predict the stability condition and SF of rock slopes prone to circular (rotational) failures.