A Machine Learning-based Predictive Model of Airspace Sector Occupancy

View Video Presentation: https://doi.org/10.2514/6.2021-2324.vid Efficient estimation of air traffic demand in airspace sectors is key to improve the overall performance of air traffic flow and capacity management strategies. This paper presents a data-driven approach for airspace sector occupancy prediction. We use supervised learning to develop prediction models to forecast sector crossing times upon aircraft takeoff from the origin airport and to estimate future sector occupancy based on operational information extracted from flight trajectory, meteorological, and traffic flow management restrictions data. Several machine learning methods are considered and assessed in terms of predictive performance: Extreme Gradient Tree Boosting, Random Forests, Support Vector Regression, and Artificial Neural Networks. The predictive models are applied to forecast peak 15-minute occupancy for two airspace sectors of the Brasilia Area Control Center in Brazil. The predictions are compared against estimates derived from a baseline demand estimation model which mimics current practice. We find that the knowledge derived from operational data contributed significantly to enhance the predictability of airspace sector demand.