The large amount of data available nowadays allow to develop powerful machine learning models, which otherwise would require laborious design of hard-coded rules by domain experts. During the first wave of InnoLabs carried out at the EUROCONTROL Innovation Hub (EIH), two models were trained on historical traffic data to solve different yet related tasks. The first model was trained to predict the evolution of the air traffic management (ATFM) delay of a flight caught by one or more ATFM regulations. The second model was trained to predict the propagation of arrival delay along the sequence of flights of an aircraft, aiming to assess its risk of curfew infringement. This presentation exposes the architecture and working principle of the models, as well as the main results and lessons learnt.

Download the slides for this talk.Download ( PDF, 4223.71 MB)