Probabilistic analysis of gravity retaining wall against bearing failure

Abstract

Machine learning (ML) models have been extensively used in the stability check of gravity retaining wall. They are renowned as the most capable methods for predicting factor of safety (FOS) of gravity retaining wall against bearing failure. In this work, FOS against bearing is predicted based on extreme gradient boosting (XGBoost), random forest (RF) and deep neural network (DNN). To establish homogeneity and distribution of datasets, Anderson-Darling (AD) and Mann-Whitney U (M-W) tests are carried out, respectively. These three machine learning models are applied to 100 datasets by considering six influential input parameters for predicting FOS against bearing failure. The execution of the established machine learning models is assessed by several performance parameters. The obtained results from computational approach shows that DNN attained the best predictive performance with coefficient of determination (R 2) = 0.998 and root mean square error (RMSE) = 0.006 in the training phase and R 2 = 0.929 and RMSE = 0.053 in the testing phase. The models result are also analyzed by using rank analysis, regression error characteristics curve, and accuracy matrix. Sensitivity analysis is carried to know the relative importance of input variables.