Training a Machine Learning Model to Detect Holding Patterns in Aircraft Trajectories

Abstract

This paper presents a Machine Learning (ML) model developed to detect holding pattern events in aircraft trajectories. Holding patterns are racetrack-shaped flight paths that an aircraft follows while awaiting further instructions or clearance from air traffic control (ATC). They are typically used to delay an aircraft’s approach or to maintain flight without progressing towards its destination, often due to airport congestion, adverse weather conditions, or other operational factors. Accurate detection of these patterns in aircraft trajectories is crucial for performance evaluation studies within Terminal Manoeuvring Areas. Although holding patterns are relatively straightforward to define, efficiently detecting them using rule-based methods is challenging. This study details the process of labelling a dataset comprising over 130,000 aircraft trajectories landing at five major European airports and training a model to accurately identify these patterns.