Embry Riddle Aeronautical University
Domain Knowledge-informed Machine Learning Model to Study Weather-induced Flight Delays
Abstract
dc:description.abstractWeather-induced Traffic Management Initiatives (TMIs) occur rarely relative to normal air traffic operations. This scarcity produces severe class imbalance that limits predictive models of flight delay. To address it, this dissertation leverages a physics- and domain- knowledge informed Variational Autoencoder (VAE), conditioned on airport identity, that generates physically plausible synthetic weather events for underrepresented TMI categories. The fidelity of the model is assessed with a calibrated discrepancy measure disclosed along- side the results rather than described in free-form terms. A multitask Graph Neural Network (GNN) is then trained on the augmented dataset to jointly classify airport-level TMI status and predict flight delays across a 39-airport network, achieving a strong, stable aggregate performance. Its remaining errors are shown to concentrate in specific, interpretable cir- cumstances rather than being distributed unpredictably across the dataset. Together, thee results show that physics-informed synthetic augmentation can improve multitask prediction of weather-driven disruption, while offering a precise , quantified account for where and why the approach still falls short.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy in Electrical Engineering & Computer Science
- Level thesis:degree_level
- Dissertation - Open Access
- Discipline thesis:degree_discipline
- Electrical Engineering and Computer Science
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bawa, Gurvir
Subjects
dc:subject × 9Identifiers
dc:identifier.*- Repository record dc:identifier
- https://commons.erau.edu/edt/1026