A Shallow Foundation Settlement Prediction Method Considering Uncertainty Based on Machine Learning and CPT Data

In the field of geoengineering, predicting foundation settlement is a critical topic. Traditional settlement prediction methods struggle to accurately reflect settlement under complex geological conditions. This study combines cone penetration test (CPT) data and collects data from 46 different geoengineering sites from the literature. Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), Deep Neural Network (DNN), Support Vector Machine (SVM), and Random Forest (RF) models are individually established, and an ensemble model is proposed to predict shallow foundation settlement St. The results show that the proposed ensemble model exhibits the best predictive performance, providing a reference for practical engineering projects. The predictions of the optimal model are compared with those of single models and traditional methods, and the uncertainty of model predictions is quantified using Monte Carlo Simulation (MCS). Sensitivity analyses are conducted using feature importance analysis and SHAP methods to assess the influence of input parameters on the prediction results. Finally, Generative Adversarial Networks (GANs) are introduced to generate new data to validate the generalization capability of the model.