Reliability Analysis of Rock Slope Using Soft Computing Techniques

Probability of safety (reliability) analysis is a major concern of any structure, especially of rock mechanics. This paper used different machine learning (ML) techniques (cubist model, extreme learning machine (ELM) and multivariate adaptive regression splines (MARS)) for reliability analysis of rock slopes. The performance of these ML models was assessed using different statistical parameters, such as Nash-Sutcliff coefficient (NS), coefficient of determination (R-2), root mean square error (RMSE), variance account factor (VAF), expanded uncertainty (U95), mean absolute error (MAE), ... etc. A comparative study was performed to test the adaptability of cubist, ELM and MARS models. It is evident from the results that MARS model shows excellent results in terms of fitness parameters. This study reflects that cubist, ELM and MARS models are well capable of predicting the reliability of slope in terms of the factor of safety (FOS) of rock slope considering statistical predictands.