Prediction of meteorological delays in terminal airspace using artificial intelligence

This thesis addresses the challenge of predicting flight delays caused by meteorological conditions at major hub airports using artificial intelligence. Utilizing data from Istanbul Airport (LTFM) as a case study, models were developed by integrating standard meteorological reports (METAR) with operational flight records. The research designed both a regression model to forecast delay durations and a classification model to assign delays into operational risk categories. The challenges posed by rare high-delay events and data imbalance were successfully overcome using advanced synthetic data generation and balancing techniques, such as CTGAN and ADASYN. The developed LightGBM and H2O AutoML-based models produced high-performance predictions on test data. Notably, the classification model demonstrated exceptional success in identifying operationally critical "High Delay" events. SHAP analyses provided transparency into the models' decision mechanisms, confirming the decisive role of factors like low visibility and low cloud ceilings. This research presents an interpretable and robust methodological framework for forecasting meteorological delays in complex aviation operations.