A Machine Learning Approach to Predict the Evolution of Air Traffic Flow Management Delay

In Europe, the most common air traffic flow management measure used by the network manager to resolve imbalances between demand and capacity is to impose regulations, which delay flights on ground. The ground delay assigned to a regulated flight may change from the time it is caught by the regulation(s) to the actual departure. This variability of the delay stems from the mechanisms used by the computer-assisted slot allocation system to manage the slots of the regulations. At present, the information on the delay evolution of a regulated flight is very limited for the airspace users, raising high uncertainty on the delay propagation and the operations management throughout the day. This paper describes the architecture of a machine learning model that, trained on historical data, is able predict the evolution of the delay for a regulated flight. Such evolution is expressed by using various indicators, which were selected by the airspace users involved in the project. The proposed model is able to predicted the trend of the delay with an accuracy of 0.75. Furthermore, results show that the model is able to reduce the prediction error (measured as the difference between the actual and the predicted delay) up to 63%, if compared to the current delay as reported by the enhanced tactical flow management system.