Shape Analysis of Flight Trajectories Using Neural Networks

Abstract

The recent widespread implementation of Automatic Dependent Surveillance–Broadcasting (ADS-B) systems on aircraft allows for improved monitoring and air traffic control management. As part of this monitoring, it is important to be able to detect unusual flight trajectories due to weather events, detection avoidance, aircraft malfunction, or other activities that may signal anomalous behavior. Given the large volume of ADS-B data available from aircraft around the world, the ability to automatically determine the shape of the trajectory and identify anomalous behavior is important to reduce the need for human identification and labeling. A neural network model is developed for multicategory classification of the shape of the trajectory using features derived from a large ADS-B data set such as bearing and curvature. The results suggest promise in differentiating common trajectory shapes using key factors, with the accuracy of the classifier being comparable to human accuracy.