Prediction of factor of safety of slopes using stochastically modified ANN and classical methods: a rigorous statistical model selection approach

Different methods like limit equilibrium and soft computing-based methods are scattered in the literature for the prediction of the factor of safety (FoS) of slopes. However, selecting reliable models among them may be difficult for the users. Therefore in this study, we propose two different hybrid ANN models and perform the reliability analysis of the existing models and the proposed models using the historical datasets. The obtained datasets comprised the geotechnical properties of the soil and the slope geometric parameters. Subsequently, the ANN models were simulated, and the optimum ANN model was selected and then subjected to two stochastic optimization algorithms to improve its performance. Next, the performance of the ordinary and hybrid ANN models was compared using the empirical cumulative frequency distribution (CFD). Thereafter, 19 independent datasets outside those used in developing the models were used to validate the proposed models, the classical slope stability analysis models along with an existing ANN model. The validation was done using both the empirical CFD and mean absolute relative error (MARE). The results in all the validation cases favored hybrid ANNs. Then, the models were further subjected to rigorous statistical analysis by subjecting the models to the normality test, analysis of variance (ANOVA), variance homogeneity test, two-way t test, and nonparametric test. The output of all the tests conducted in this study revealed that the hybrid ANNs are most suitable for the slope stability analysis.