Predicting Deviation of Flight Entry into Air Sector using Machine Learning Techniques

Abstract

The management of air traffic is a complex task that requires ensuring the safety and efficiency of aircraft trajectories when transiting from one airspace sector into another. This work explores the use of historical flight data to predict if a flight will commit to the planned entry point when entering an airspace sector. To achieve this, we propose a feature engineering method that can be employed to convert raw flight data into a matrix which captures flight count information in predefined grids. This matrix is referred to as the Air Space Occupancy Grid (ASOG) and it captures the state of traffic in an airspace sector and its immediate vicinity. Experiments are performed using the Swedish Civil Air Traffic Control (SCAT) dataset. To predict whether an aircraft will deviate from its planned entry point, supervised machine learning algorithms are used to train a model. Through experiments on real-world data, we showcase that ASOG provides a systematic way of incorporating the state of the airspace sector and improving the performance of prediction models compared to simple features. The prediction output can be used to notify human air traffic controllers in advance about potential deviation to flight plan upon entry to an airspace sector. This can improve the planning process of air traffic controllers in their work in maintaining safe and efficient air traffic.