EUVIMEDEUVIMED entdecken
Crossref

Machine-Learning Estimation of Clay Undrained Shear Strength at New Sites: A Critical Review and Site-Held-Out Benchmark

Mustafa S. Alsumaidaie, Mohamed M. Al-Heeti, Ahmed K. Hammood

Journal of Al-Farabi for Engineering Sciences · 01.12.2025 · 4 (2) · 373-383

DOI: 10.59746/h0hf9s77 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Al-Farabi for Engineering Sciences, online erschienen am 01.12.2025. Verfasst von Mustafa S. Alsumaidaie und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine-learning models are increasingly used to estimate the undrained shear strength of clay from index properties, stress history and penetration tests, and several report coefficients of determination above 0.9. This paper reviews how such models have been validated and what uncertainty they report, then tests the main concern it raises on a public global database. Sixteen data-driven strength studies published between 2012 and 2025 were screened and coded. Eight evaluated their models only on randomly split records, six did not describe the protocol in the material available, and two tested at sites withheld from training. Only one reported prediction intervals with an empirical coverage check. The benchmark used 2,340 records from 325 sites and compared linear, random forest and gradient-boosting models with the SHANSEP and Mesri relations under random, site-held-out and country-held-out cross-validation. For gradient boosting, the root mean square error of the logarithm of the normalised strength rose from 0.147 under random splitting to 0.175 at unseen sites and 0.195 in unseen countries, and its apparent advantage over SHANSEP shrank from 0.036 to 0.009 before disappearing. Split-conformal intervals at a nominal 90% level, calibrated on random records, covered 84.9% of records at unseen sites and 80.4% in unseen countries. Calibrating on whole sites brought coverage back to the nominal level at unseen sites; unseen countries also needed equal site weights, and the intervals became wider. Strength models meant for new sites should therefore be validated and calibrated on whole sites, following the proposed reporting checklist.

Zitieren

APA

Alsumaidaie, M. S., Al-Heeti, M. M., & Hammood, A. K. (2025). Machine-Learning Estimation of Clay Undrained Shear Strength at New Sites: A Critical Review and Site-Held-Out Benchmark. Journal of Al-Farabi for Engineering Sciences, 4(2), 373-383. https://doi.org/10.59746/h0hf9s77

Vancouver

Alsumaidaie MS, Al-Heeti MM, Hammood AK. Machine-Learning Estimation of Clay Undrained Shear Strength at New Sites: A Critical Review and Site-Held-Out Benchmark. Journal of Al-Farabi for Engineering Sciences. 2025;4(2):373-383. doi:10.59746/h0hf9s77

BibTeX
@article{Alsumaidaie2025,
  author = {Alsumaidaie, Mustafa S. and Al-Heeti, Mohamed M. and Hammood, Ahmed K.},
  title = {Machine-Learning Estimation of Clay Undrained Shear Strength at New Sites: A Critical Review and Site-Held-Out Benchmark},
  journal = {Journal of Al-Farabi for Engineering Sciences},
  year = {2025},
  volume = {4},
  number = {2},
  pages = {373--383},
  doi = {10.59746/h0hf9s77}
}
Kostenlos starten →