Predicting Undrained Shear Strength From Towed-TEM Resistivity And Sparse CPT Data Using Machine Learning Models
Abstract
Undrained shear strength (Su) is a critical geotechnical parameter for foundation design, particularly in embankment structures. Traditional field and laboratory methods for estimating Su are spatially limited and costly. This study presents a machine learning (ML) workflow for predicting Su using electrical resistivity data from the towed transient electromagnetic method (tTEM) along the Kaskaskia Levee. Since ML models require large datasets, which geotechnical datasets often lack to reliably train model parameters, we introduce generative adversarial networks (GANs) to augment the original dataset by generating a catalog of 20,000 synthetic datapoints. Three ML algorithms, including Support Vector Regression (SVR), Random Forests (RF), and Extreme Gradient Boosting (XGBoost) were trained on both the original and synthetic datasets. We evaluated model performance using three feature sets: depth only, resistivity only, and depth plus resistivity. Results show that the best performance is obtained when including electrical resistivity and depth, although depth was the most important prediction parameter. Models trained on the larger synthetic data set showed better performance than those trained only on the original data. This result highlights the benefit of data augmentation in overcoming limitations of sparse field data. Predicted Su values along the Kaskaskia Levee shows Su ranging from ∼300 kPa near the surface to ∼1500 kPa at ∼15 m depth. Variability along the length of the profile, which is especially significant at depths beyond 10 m, is constrained by the electrical resistivity observations. This workflow offers a method for predicting Su and potentially other geotechnical parameters from sparse subsurface data.
Recommended Citation
K. Arowoogun et al., "Predicting Undrained Shear Strength From Towed-TEM Resistivity And Sparse CPT Data Using Machine Learning Models," Journal of Applied Geophysics, vol. 254, article no. 106500, Elsevier, Nov 2026.
The definitive version is available at https://doi.org/10.1016/j.jappgeo.2026.106500
Department(s)
Geosciences and Geological and Petroleum Engineering
Keywords and Phrases
CPT; GAN; Levee; Synthetic data; tTEM; Undrained shear strength
International Standard Serial Number (ISSN)
0926-9851
Document Type
Article - Journal
Document Version
Citation
File Type
text
Language(s)
English
Rights
© 2026 Elsevier, All rights reserved.
Publication Date
01 Nov 2026
Comments
Missouri University of Science and Technology, Grant None