Pre-tactical prediction of ATFM delay for individual flights

Airlines develop an operation plan, during the day prior to operations (D-1), to identify potential network issues and prepare potential pre-tactical preventing measures such as aircraft tail swapping or crew reassignment to be applied on D0. Flights might experience discrepancies between their plan and execution due to many different factors, and in particular demand-capacity imbalances in the network leading to Air Traffic Flow Management (ATFM) regulations. Dispatcher3, a Clean Sky 2 innovation action, focuses on the use of machine learning techniques to support the airlines processes prior departure: dispatching, understood as the broad flight planning from the day prior to operations to the flight plan definition and selection, and advisories to pilot. This paper focuses on the estimation of ATFM delay for individual flights during the pre-tactical phase (D-1), which could help airspace users apply mitigation actions. Four machine learning models are developed to produce individual independent estimations with different level of granularity. The first two are binary classifier models that provide information on the probability of a given flight being affected by ATFM delay, and the reason for this delay (airport or airspace congestion). These models reported an accuracy between 75% and 88%. The later two models estimate the impact of the delay (amount of delay assigned to the flight if regulated), with a Mean Absolute Error close to 9.35 minutes.