A Two-Stage Uncertainty-Aware Approach to Predict Flight Delays in the European Airspace

This working paper presents a two-stage, uncertainty-aware predictive framework for flight delays across the European airspace. The study addresses three key gaps in the existing literature: the scarcity of research at the European scale, the limited integration of en-route meteorological information, and the under-explored quantification of predictive uncertainty. We propose a modelling approach that combines operational flight data from Eurocontrol with ERA5 reanalysis weather data extracted along entire flight trajectories. Departure delays are predicted using gradient-boosted decision trees, and these predictions are subsequently incorporated as features in the arrival delay model. To provide calibrated prediction intervals, we employ Conformalised Quantile Regression. Model performance is evaluated using temporal cross-validation and an out-of-sample test on 2023 data, demonstrating competitive predictive accuracy and reliable uncertainty quantification. This text is a preprint. As such has not yet been peer-reviewed and may undergo substantial changes.