Comparative machine learning study for allowable bearing capacity prediction of OHTL tower foundations with spatially imputed geotechnical borehole databases’ interpretability
Challenge Journal of Structural Mechanics · 24.06.2026 · 12 (2) · 125
DOI: 10.20528/cjsmec.2026.02.006 ↗Kurz gefasst
Fachartikel in der Zeitschrift Challenge Journal of Structural Mechanics, online erschienen am 24.06.2026. Verfasst von Esra Uray und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Geotechnical site characterization for alignment-based infrastructure projects, such as overhead transmission lines (OHTLs), involves the collection of extensive borehole datasets across hundreds of kilometers and the interpretation of thousands of results from laboratory and in-situ tests. In this study, a machine learning (ML) framework is proposed for predicting the allowable bearing capacity (qₐₗₗ, kPa) of OHTL tower spread footings using geotechnical borehole data. A dataset comprising 89 boreholes with 16 input parameters, including physical, mechanical, and seismic soil properties, was used. Missing borehole records (n=8) were spatially imputed using Inverse Distance Weighting (IDW) and K-Nearest Neighbors regression prior to model training. Ten ML algorithms, including Ridge, Lasso, ElasticNet, Support Vector Regression, K-Nearest Neighbors, Random Forest, Extra Trees, Gradient Boosting, XGBoost, and LightGBM, were trained. Model performance was assessed using coefficient of determination (R²), root mean square error, mean absolute error, and mean absolute percentage error metrics. Random Forest was identified as the most reliable model for practical deployment, exhibiting balanced generalization behavior with training R² converging from 0.63 to 0.98 and a Train-CV Gap of 0.017, confirming that it learns generalizable patterns rather than memorizing individual observations. Although Gradient Boosting achieved the highest overall performance metrics, its persistent Train R² of 1.000 indicated memorization behavior. SHAP-based interpretability analysis identified groundwater depth, shear strength parameters, and standard penetration test blow count as the primary effective parameters of the qₐₗₗ value. The proposed framework shows that ML-based approaches can significantly enhance the reliability of bearing capacity predictions along transmission line corridors.
Zitieren
APA
Uray, E., & Yildiz, T. (2026). Comparative machine learning study for allowable bearing capacity prediction of OHTL tower foundations with spatially imputed geotechnical borehole databases’ interpretability. Challenge Journal of Structural Mechanics, 12(2), 125. https://doi.org/10.20528/cjsmec.2026.02.006
Vancouver
Uray E, Yildiz T. Comparative machine learning study for allowable bearing capacity prediction of OHTL tower foundations with spatially imputed geotechnical borehole databases’ interpretability. Challenge Journal of Structural Mechanics. 2026;12(2):125. doi:10.20528/cjsmec.2026.02.006
BibTeX
@article{Uray2026,
author = {Uray, Esra and Yildiz, Tahir},
title = {Comparative machine learning study for allowable bearing capacity prediction of OHTL tower foundations with spatially imputed geotechnical borehole databases’ interpretability},
journal = {Challenge Journal of Structural Mechanics},
year = {2026},
volume = {12},
number = {2},
pages = {125},
doi = {10.20528/cjsmec.2026.02.006}
}
EUVIMED