Application of MCS, GRNN, and GPR for performing the reliability analysis of rock slope

The failure of rock slopes leads to disastrous consequences and thus necessitates their reliability analysis. There are various methods to perform the reliability analysis of a rock slope, viz., conventional, numerical, and Soft Computing (SoCom). However, due to different environmental and loading conditions, there is a need to explore SoCom methods to evaluate the probability of failure of rock slopes. Therefore, in this paper, three soft computing techniques viz., Monte Carlo Simulation (MCS), Generalized Regression Neural Networks (GRNN), and Gaussian Process Regression (GPR) have been proposed to analyze the rock slope stability under various conditions. The variables used in this study were c, ϕ, and σt, and the output was the factor of safety (Fs), which was further utilized for the training and testing of the models. The training and testing of the models were performed on different sets of datapoints (viz., 50, 100, 300, 500, 1000, 5000, 10,000). The MCS algorithm was used to generate various sets of datapoints. Furthermore, statistical parameters were used to assess the performance of the proposed SoCom. A comparative study has been performed to check the adaptability of MCS, GRNN, and GPR models for performing the reliability analysis of rock slopes. Training vs. testing datasets has been plotted to realize the fitting of these models. Furthermore, the observations from Taylor’s plot indicate that the MCS, GRNN, and GPR models are capable of predicting the reliability of slope in terms of reliability index; however, GPR outperformed the other two models. The findings imply that the performance assessment of MCS, GRNN, and GPR should also be tried for reliability and risk analysis of other geotechnical and rock engineering problems.