Flight delays are a well-known source of concern for the air traffic management. Numerous assessments of delay causes have revealed that reactionary delay contributes the most to the average delay per flight. The reactionary delay has several causes, including awaiting load and/or cabin crew from another flight, as well as the late arrival from the inbound flight operated by the same aircraft, also known as rotational reactionary delay. While the former is likely unpredictable without access to load and cabin crew schedule information, which airlines do not publish, rotational reactionary delay could be predicted to some extent using data available to airports and the Network Manager (In the context of this study, the term ‘Network Manager’ refers to a key entity within the air traffic management system. The Network Manager’s role varies across regions, but generally, it is responsible for coordinating and optimizing the use of airspace and airport resources to ensure efficient air traffic flow. This role is particularly crucial in Europe, where the Network Manager is tasked with managing a complex and highly interconnected airspace. Please note that the interpretation of ‘Network Manager’ may differ depending on the reader’s regional context). This paper proposes a tree-based model for predicting the rotational reactionary delay of individual flights that combines the most recent advances in natural gradient-boosting and feature attribution methods to provide predictions that are both probabilistic and explainable. The effectiveness of the model is demonstrated for two distinct yet related quantities: departure delay distribution (a probabilistic regression task) and likelihood of being punctual at departure (a binary classification task). In both cases, the model is trained and tested using two distinct datasets containing roughly the same information: operational data from the enhanced tactical flow management system, and the research and development data archive made available to the public by EUROCONTROL. Results show that the proposed model outperforms predictions based on standard statistical methods. Furthermore, findings evidence the importance of a precise problem statement when training machine learning models on uncertain flight data.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Probabilistic and explainable tree-based models for rotational reactionary flight delay prediction


    Additional title:

    CEAS Aeronaut J


    Contributors:

    Published in:

    CEAS Aeronautical Journal ; 15 , 4 ; 1157-1173


    Publication date :

    2024-10-01


    Size :

    17 pages




    Type of media :

    Article (Journal)


    Type of material :

    Electronic Resource


    Language :

    English





    Explainable Multi-task Flight Arrival Delay Prediction

    Wang, Tianqi / Lin, Lei / Gao, Jing | IEEE | 2021


    Flight Evaluation of Reactionary Envelope Protection System for UAVs

    Unnikrishnan, S. / Prasad, J. V. R. / American Helicopter Society | British Library Conference Proceedings | 2006


    Lap expanding reactionary airbag

    AKOMA PATRICK / MARROQUIN MANUEL / GARCES KASEY | European Patent Office | 2024

    Free access