Predicting the likelihood of airspace user rerouting to mitigate air traffic flow management delay. (November 2022)

• Record Type: Journal Article
• Authors: Dalmau, Ramon
• Is Part Of: Transportation research. Volume 144(2022)
• Publication Date: 2022-11
• Subjects: Air traffic flow management -- Flight delay -- Machine learning -- Model interpretability; Transportation -- Periodicals; Transportation -- Technological innovations -- Periodicals; 388.011
• DOI: 10.1016/j.trc.2022.103869
• Languages: English
• ISSNs: 0968-090X

Abstract

Abstract: At present, the most common air traffic flow management measure used by the European Network Manager to solve en-route demand and capacity balance issues is to impose air traffic flow management regulations in congested airspace, which delay flights on the ground to smooth demand. The delay imposed on a regulated flight, however, has a negative operational and economic impact on the corresponding airspace user. When the impact is severe, the airspace user may reroute the flight to tactically circumvent the regulated airspace, thereby avoiding the regulation and reducing (or even eliminating) the delay. This legitimate strategy is unquestionably effective in mitigating the delay and the associated cost, but it may also increases traffic demand uncertainty due to last-minute changes as well as the environmental impact due to sub-optimal trajectories. This paper presents a gradient-boosted decision trees model that, trained on historical data, could assist in predicting the likelihood of a regulated flight rerouting to mitigate air traffic flow management delay. Specific metrics for highly imbalanced binary classification problems reveal that, while not perfect, the model outperforms a rudimentary decision tree and a dummy classifier. Furthermore, Shapley values computed with the model uncover the most important triggers for tactical rerouting from a data-driven perspective, namely the current delay, the aircraft operator, the airspace sector where the most penalising regulation is applied and the remaining time until departure. This interpretable model could be used to assist flow managers in their decision-making, as well as to accurately mimic the behaviour of flight dispatchers when facing delays in simulators. Highlights: Machine learning can predict the likelihood of flight rerouting to mitigate delay. The importance of the input features reveal the strategy of the airspace users. The most important factors are the current delay and the aircraft operator.