Data-driven prediction of aircraft holding times using OpenSky data

Abstract

As global air traffic increases, major hubs such as London Heathrow and Gatwick experience increasing congestion, leading to the frequent queuing of aircraft in holding stacks. This study employs machine learning (ML) techniques to predict aircraft holding times at London Heathrow, aiming to enhance flight management and reduce congestion in the Terminal Maneuvering Area (TMA) by enabling adjustments to flight trajectories and speeds earlier in the flight, rather than upon arrival. Leveraging historical data - including surveillance data from the OpenSky Network - from April 2023 to March 2024, the study adopts the LightGBM and LSTM ML frameworks to develop two sets of predictive ML models: one set of regression models to predict the holding time of individual flights up to 60 minutes from the London TMA; and one set of time series regression models to predict the average holding time in different holding stacks. Test results of the regression models show that the models trained with LightGBM have the best performance, with minimum RMSE and MAE values of 2.25 and 1.50 minutes, respectively. On the other hand, the results of the time series regression models show better performance by the models trained with LSTM, with average RMSE and MAE values of 2.41 and 1.47 minutes, respectively. In conclusion, this research highlights the effectiveness of ML in predicting aircraft airborne holding times, offering significant benefits to pilots, air traffic controllers, and flight planners. The successful application of these models could lead to substantial improvements in flight efficiency and reduced environmental impact from fewer delays and less fuel consumption.