Neural network based air traffic controller workload prediction

This paper develops a method which relates traffic data to workload situation in qualitative terms such as high, medium and low. The method is based on earlier research in image processing and pattern recognition that has identified measures of second-order statistics which characterize different regions within the image based on gray level patterns. For characterizing air traffic patterns, these measures have been computed using position and velocity of the aircraft within the airspace. In order to relate the patterns in terms of their statistics to the controller's assessment of the workload, a multilayer neural network has been used. Recorded air traffic data using the Center TRACON Automation System and controller's rating of the same data, have been used to train the neural network. It is shown that a trained neural network with statistical measures derived from the traffic as input can be used for predicting air traffic controllers workload that is in agreement with the qualitative assessment of their workload.