Accuracy of machine learning techniques in forecasting tunnelling-induced soil settlements with limited data

1. Home
2. Publications
3. Online Library
4. Accuracy of machine learning techniques in forecasting tunnelling-induced soil settlements with limited data

Publication Title
Accuracy of machine learning techniques in forecasting tunnelling-induced soil settlements with limited data

Abstract
This article presents a comprehensive comparison of the predictive accuracy of 5 Machine Learning algorithms – Support Vector Machine, Decision Tree, Random Forest, Extreme Gradient Boosting (XGBoost) and Back- propagation Neural Network – in forecasting maximum settlements induced by earth pressure balance tunnel boring machines. The study evaluates those algorithms’ performance using a dataset of 2,590 settlement measurements associated with soil parameters and shield operational parameters, collected from 13 km of the Grand Paris Express Project (line 15 South West and line 14 South). The analysis reveals that, for the small dataset used in this study, ensemble methods, specifically XGBoost and Random Forest, outperform artificial neural networks in predicting settlements. These methods remain reliable even with only 30% of the dataset used for training. The article further explores the regularization and optimization of hyperparameters for Random Forest and XGBoost, shedding light on their enhanced predictive capabilities. Furthermore, variations in the data split between training and testing sets are explored, highlighting the effectiveness of these Machine Learning algorithms in achieving accurate predictions with limited datasets.
Authors
T. Richa; Jean-Michel Pereira; L. M. Guayacan-Carrillo; Gilles Chapron; F. Lanquette

Published In

18th European Conference on Soil Mechanics and Geotechnical Engineering (ECSMGE2024)

Theme

C - Risk analysis and safety evaluation

Download PDF