Aircraft Categorization Approach Using Machine Learning to Analyze Aircraft Behavior

Abstract

The establishment of aircraft categories is a classification technique employed in a variety of aviation disciplines, including design and development, certification, ongoing airworthiness, air traffic management, surveillance, and safety analysis. Traditional approaches rely on manual feature engineering, which can be labor-intensive and ineffective for capturing complex patterns. In this paper, an approach to aircraft categorization using unsupervised machine learning clustering is proposed. The aim of the proposed approach is to be simple in order to be useful and understandable across disciplinary domains; to be scalable to large volumes of air traffic data; and to be adaptable to changes to account for the evolving technological and operational nature of the airspace environment. The application is based on an adapted version of the k-means algorithm that can group aircraft into clusters based on 3D position over time. The approach is validated using real-world, publicly available ADS-B air traffic data, and the results are compared to traditional categorization methods from the field of aircraft certification. The results showed that the model could be used to 1) identify and group aircraft sharing the same flight phase, 2) categorize aircraft with a similar general heading or direction, and 3) distinguish between local regional aircraft operations and longer flight operations. It was also shown that, depending on the use case, the model could be extended to identify more granular behaviors by increasing the K value used to create the model. Overall, the findings demonstrate that leveraging machine learning techniques for aircraft categorization provides an effective, automated, and scalable solution applicable to a wide range of current applications.