Internal Stability of Mechanically Stabilized Earth Wall Using Machine Learning Techniques

This paper proposes an AI-based prediction method for factor of safety (FOS) against rupture and pull-out failure and examines and compares the applicability and adaptability of k -nearest neighbor (KNN), random forest (RF), and extreme gradient boosting (XGBoost) in the reliability analysis of internal stability of mechanically stabilized earth wall. In order to predict FOS against rupture and pull-out failure, these three machine learning (ML) models are applied to 100 datasets, taking into account three crucial input parameters, namely, the depth of the reinforcement layer below the wall’s crest, the angle of internal friction of soil, and the unit weight of soil. A variety of performance indicators, including coefficient of determination ( R 2 ), variance account factor, Legate and McCabe’s index, A-10 index, root mean square error (RMSE), expanded uncertainty, mean absolute error, and median absolute, are engaged to assess the effectiveness of the well-established ML models. The results, which are based on performance metrics, indicate that XGBoost outperformed in both the two cases, i.e., rupture and pull-out, than the other proposed machine learning models in terms of predictive performance. This can be attributed to its highest R 2 = 0.999 and lowest RMSE = 0.002 (in case of rupture) and highest R 2 = 0.999 and lowest RMSE = 0.003 (in case of pull-out failure) during the training phase, as well as its highest value of R 2 = 0.864 and lowest value of RMSE = 0.102 (in case of rupture) and R 2 = 0.944 and RMSE = 0.054 (in case of pull-out failure) during the testing phase. Rank analysis, reliability analysis, regression plot, confusion matrix, and error matrix plot are the other tools used to assess the performance of the model. Using first-order second moment methods, the reliability index of the model is computed and compared to the actual value for both cases. Sensitivity analysis is also performed to regulate the effect of each input parameter on both outputs.