Task-Based Machine Learning Model for Terminal Air Traffic Complexity Prediction

Air traffic complexity is the degree of difficulty in monitoring and managing a specific air traffic situation, and it is one of the main drivers of Air Traffic Controller (ATCO) workload. Previously developed terminal air traffic complexity models and methods require significant adjustment or additional data to be applied to unseen airspaces, or to be compared with ATCOs’ complexity perception. To address this gap, a terminal air traffic complexity model based on approach ATCO tasks was developed. Building on previous work, approach ATCO tasks were quantified and used as input features. A new dataset consisting of three different terminal airspaces, each containing a different number of situations, was created. Certified approach ATCOs evaluated the complexity of each situation on a scale of 1 to 5, and these values were used as ground truth for training ordinal regression models. One linear and one non-linear model were trained and evaluated, and both were tested on an airspace unseen during training. The two models achieved strong performance on the tested airspace with Quadratic Weighted Kappa (QWK) values of 0.814 and 0.873, confirming that the proposed task features can predict the complexity of a new, unseen airspace.