Unveiling airline preferences for pre-tactical route forecast through machine learning

In this work we describe a novel approach for the prediction of the flight plan to be sent by airspace users during the pre-tactical phase of Air Traffic Flow and Capacity Management (ATFCM). The proposed approach uses machine learning algorithms to extract airspace user preferences in terms of route characteristics, allowing the prediction of new routes not observed during the model training phase. We present the results obtained from applying this approach to short and medium range KLM flights for 52 weeks. Results show that the proposed solution is robust, scalable and capable of reducing the number of wrong predictions provided by the current Network Manager operational solution by 24.3% (4.5% increment on accuracy).