Enhancing ground modeling through artificial intelligence

Details

Supervisors

Rattez, Hadrien ; Vanden Berghe, Jean-François

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en sciences informatiques, à finalité spécialisée

Abstract

enThis thesis explores how machine learning can enhance ground modeling by integrating geophysical and geotechnical data. Two case studies were analyzed: a regression task predicting continuous CPT parameters (e.g., cone resistance, friction) from geophysical measurements, and a classification task assigning soil types based on geophysical profiles. The study demonstrates that CatBoost, the ML algorithm used, can extract useful patterns even from sparse and noisy datasets. However, performance varies significantly with data quality, geological complexity, and spatial configuration. Key findings include the importance of spatially consistent validation, the benefits of class grouping for improving classification accuracy, and the limited impact of sample weighting strategies. The thesis also highlights the need for hybrid models that incorporate domain knowledge, as well as the ethical and practical considerations for real-world deployment. Overall, this work shows that machine learning, when carefully applied, can become a valuable support tool for subsurface characterization and engineering decision-making.