Application of LS-SVM Method in Probabilistic Stability Analysis of Saturated Soil Slopes

To assess the stability of a soil slope considering uncertainty in soil characteristics (C, Φ), it is necessary to conduct repeated calculations to determine both the Factor of Safety (FS) and the Probability of Failure (PF). The absence of analytical solutions leads researchers to use surrogate approaches of probabilistic modeling. This study aims to assess the PF and FS of a saturated soil slope using one of the Machine Learning (ML) method subsets, the regression Least Squares Support Vector Machine (LS-SVM). The results are compared to those obtained from the Limit Equilibrium Method (LEM) to compare the accuracy of LS-SVM. According to the FS-PF plot, it is concluded that LS-SVM is a reliable tool for determining the stability of a multi-layered saturated soil slope, which significantly improves computational efficiency. Furthermore, since predicting FS and PF from field data requires fewer data points, it will result in significant time and cost savings. However, to avoid false predictions, limitations must be removed, for instance, each layer should be included in the analysis within its probabilistic distribution, so using this method is more efficacious in clay soils where there are values for cohesion, unlike sands (c=0). This helps training data to be prepared well and to prevent misprediction in local failures. In addition, training data are debated in terms of the percentage employed. Furthermore, showing the importance of dataset scattering is concluded that adopting data with uniform distributions plays a crucial role in preparing training data sets and consequently, in having a better prediction of the possible slope failures.