A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Geotechnical and Earth Resources Engineering

Thesis (M. Eng.) - Asian Institute of Technology, 2025

Accurate stratigraphic classification is fundamental in geotechnical engineering, as it forms the basis for understanding subsurface conditions and developing safe, reliable, and cost-effective design solutions. However, subsurface soil and rock strata are inherently heterogeneous and spatially variable, while borehole data are typically sparse, incomplete, and noisy. Conventional subsurface modeling and visualization often rely on subjective engineering interpretation, providing limited quantification of uncertainty.This study presents an evaluation of an artificial intelligence (AI)-based method for geotechnical stratigraphy classification, focusing on subsurface profile visualization using machine learning (ML) techniques. A Random Forest (RF) classifier was developed and trained using limited borehole data, incorporating spatial coordinates, elevation, and soil layer thickness to predict lithology classes and generate one-dimensional (1D) and two-dimensional (2D) subsurface profiles. The AI-based predictions were compared with results from conventional kriging and manual interpretation to assess the performance of AI-based modeling in terms of accuracy, stratigraphic consistency, and uncertainty representation.The evaluation results indicate that the RF model outperformed conventional methods, achieving higher classification accuracy and improved consistency between predicted and observed stratigraphy, even with a limited number of boreholes. These findings demonstrate that ML-based approaches, particularly the RF algorithm, provide a robust, data-driven, and objective framework for geotechnical stratigraphic classification.Overall, this study highlights the potential of AI-based methods to enhance the reproducibility and reliability of subsurface modeling, reduce subjectivity in geological interpretation, and enable quantifiable uncertainty estimation offering a practical and efficient alternative for modern geotechnical engineering applications.