Events

PhD Dissertation Defense: Estimating Pile Bearing Capacity Using Machine Learning

Lecture / Panel
 
For NYU Community

Portrait of PhD candidate Baturalp Ozturk

Baturalp Ozturk, a PhD candidate in Civil Engineering, will defend the dissertation “Estimating Pile Bearing Capacity Using Machine Learning.” Faculty and students are invited to attend in person or join on Zoom.

The axial capacity of driven piles is still estimated in practice with empirical design methods that were calibrated decades ago on small collections of load tests and that carry large and systematic errors. This dissertation examines whether machine learning (ML), trained on large databases of full-scale static load tests, can predict pile capacity more accurately and more reliably than these methods, and how such models should be developed so that their reported performance can be trusted.

The work begins by assembling the load test databases and establishing performance baselines for widely used traditional design methods, then shows that even simple, untuned algorithms can match the best traditional method. Systematic feature selection identifies the pile and soil properties that carry predictive information; ensemble methods and large-scale hyperparameter optimization deliver consistent gains; and a controlled comparison of training workflows shows how improper separation of model optimization and evaluation inflates reported accuracy, leading to best practices for developing and reporting ML models in geotechnical engineering. The dissertation then compares optimized neural networks with gradient-boosted trees, benchmarks interpretable architectures (TabNet and Kolmogorov-Arnold networks), and evaluates large language models as capacity estimators.

Across all stages, properly developed ML models matched or outperformed the traditional design methods, with markedly less scatter and far fewer grossly mispredicted piles. The findings support data-driven methods as a credible complement to current pile design practice, provided that the models are developed and evaluated with discipline.

Candidate: Baturalp Ozturk, PhD Civil Engineering
Dissertation adviser and committee chair: Prof. Magued Iskander
Guidance committee: Prof. Mohsen Hossein, Prof. Luis Ceferino, Prof. Scott Brandenberg, Prof. Tarek Abdoun