A deep learning approach to flight delay prediction

• Published in: IEEE/AIAA Digital Avionics Systems Conference pp. 1 - 6
• Main Authors: Young Jin Kim, Sun Choi, Briceno, Simon, Mavris, Dimitri
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2016
• Subjects: Airports; Atmospheric modeling; Computational modeling; Computer architecture; Delays; Machine learning; Meteorology
• ISSN: 2155-7209
• DOI: 10.1109/DASC.2016.7778092

Deep learning has achieved significant improvement in various machine learning tasks including image recognition, speech recognition, machine translation and etc. Inspired by the huge success of the paradigm, there have been lots of tries to apply deep learning algorithms to data analytics problems with big data including traffic flow prediction. However, there has been no attempt to apply the deep learning algorithms to the analysis of air traffic data. This paper investigates the effectiveness of the deep learning models in the air traffic delay prediction tasks. By combining multiple models based on the deep learning paradigm, an accurate and robust prediction model has been built which enables an elaborate analysis of the patterns in air traffic delays. In particular, Recurrent Neural Networks (RNN) has shown its great accuracy in modeling sequential data. Day-to-day sequences of the departure and arrival flight delays of an individual airport have been modeled by the Long Short-Term Memory RNN architecture. It has been shown that the accuracy of RNN improves with deeper architectures. In this study, four different ways of building deep RNN architecture are also discussed. Finally, the accuracy of the proposed prediction model was measured, analyzed and compared with previous prediction methods. It shows best accuracy compared with all other methods.