A Machine Learning-Based Extraction of Cruise Phase from Trajectory Data

When analyzing historical trajectory data, identifying the cruise phase is crucial. However, due to various vertical maneuvers along the course of the flight, it is not straightforward to extract the cruise phase, and rule-based algorithms tend to be inaccurate when the vertical trajectory becomes complicated. This study presents a machine learning-based technique to extract the cruise phase of a flight from recorded trajectory data. The trajectory data are normalized by maximum time and altitude, and then grouped into clusters using an agglomerative hierarchical clustering technique and a Gaussian Mixture Model algorithm. Finally, a rule-based selection criterion is applied to each cluster centroid to identify search regions for top-of-climb (TOC) and top-of-descent (TOD). The TOC and TOD are extracted for the individual trajectory that belongs to the cluster. The study classified a total of 38,051 trajectories into 41 clusters. The cruise phase was extracted for 36,712 flights, accounting for 96.5% of the total trajectories. The proposed method is particularly useful when the cruise phase needs to be extracted for a large data only with the trajectory data without the flight management system information.