Application of XGBoost for Predicting Bearing Capacity of Open Caissons

Abstract

This article This study presents an analysis of the bearing capacity of open caisson foundations embedded in dense sand, employing Bolton’s failure criterion. The analysis is conducted using the finite element limit analysis (FELA) method, incorporating both upper and lower bound theorems, to gain insight into the behavior of dense sand. Bolton's model can be used to investigate the relationship between strength and dilatancy to understand the behavior of dense sand and open caisson foundations. The accuracy of the study is confirmed by comparing the findings of this analysis with several past studies. Furthermore, the XGBoost model, an ensemble learning technique that aggregates predictions from multiple decision trees, is employed to enhance prediction accuracy. In each iteration, the trees attempt to correct the errors of the preceding trees, thereby improving the reliability of the model in predicting the bearing capacity of caisson foundations. The accuracy of using the XGBoost model is multiple weak learning algorithms are trained, and a strong learner algorithm is used as a final model, including the coefficient of determination (R²), mean absolute error (MAE), and root mean squared error (RMSE), with values of 0.99, 1.61, and 10.03, respectively. These data show that XGBoost's predictions are usually similar to the bearing capacity results using the FELA approach.