Determination of the Factor of Safety against Sliding of Finite Slopes Using Classical Regression and Soft Computing Approaches

Determining the factor of safety against sliding of slopes in engineering projects is a major challenges for civil engineers. A method that can provide an accurate estimation of sliding likelihood can be a significant aid to designers. In the first part of this study, formulae based on classical regression methods such as multiple linear regression (MLR), multiple non-linear regression (MNLR), and multivariate adaptive regression splines (MARS) to calculate the factor of safety (\(\overline{{F }_{s}}\) _LEM) of finite slopes are developed. In the second part, in order to develop soft computing methods for estimating \(\overline{{F }_{s}}\) _LEM, from soft computing methods (boosted trees (BT) and gene expression programming (GEP)) and two regression methods (MLR and MNLR) data-driven based methods are used. Values of \(\overline{{F }_{s}}\) _LEM for development of classical regression and soft computing models are generated using the limit equilibrium methods (LEMs). To assess the performance of the proposed models, different statistical metrics such as R², RMSE, RE%, MAE and NSE, and graphical diagrams such as scatter plots, box plots, RE% plots and Taylor plots are used. Classical regression methods indicate that the results obtained from the MARS model is closer to the extracted results of the MNLR model. Moreover, the results showed that the performance of the GEP model with R² = 0.994, RMSE = 0.0381, RE% = 1.66%, MAE = 0.027 and NSE = 0.992 is better than the other soft computing models for estimating \(\overline{{F }_{s}}\) _LEM. Designers of simple slopes with homogenous and dry soils could consider using the proposed approaches as an alternative to traditional stability charts and limit equilibrium methods (LEM).