Aircraft Fuel Burn Estimation: The EUROCONTROL PRC 2025 Data Challenge

Abstract

The EUROCONTROL Performance Review Commission launched a data challenge in 2025 for machine-learning-based fuel burn prediction, in collaboration with OpenSky Network and TU Delft. This paper describes the dataset built for the challenge. It pairs ACARS fuel telemetry, crowdsourced through airframes.io, with ADS-B trajectory data from the OpenSky Network, augmented with flight list information from EUROCONTROL over the period April to October 2025. Consumer-grade ACARS receivers supply fuel-on-board reports at irregular intervals; ADS-B provides dense kinematic trajectories at sub-second resolution. We fuse these two sources and validate fuel labels against physics-based predictions from TU Delft's OpenAP model to infer ambiguous reporting units and filter erroneous records. The resulting training set is approximately 5 GB and includes real-world noise, coverage gaps, and operational variability. We describe the data collection pipeline, the unit inference methodology, quality assurance procedures, and the structure of the released dataset.