DICLERGE: Divide-Cluster-Merge Framework for Clustering Aircraft Trajectories

Whether descriptive, predictive or prescriptive, most analytics applications pertaining to aircraft trajectory data require clustering to group similar trajectories and discovering a representative trajectory for all as a single entity. During the process, considering a trajectory as a whole may mislead, resulting in overfitting and a failure to discover a representative trajectory. In this paper, we describe a novel clustering framework; divide-cluster-merge, DICLERGE, for the aircraft trajectory data, that divides trajectories into three major flight phases: climb, enroute, and descent. It clusters each phase in isolation, then merges them together. Our unique approach also discovers a representative trajectory, the model for the entire trajectory set. Our experiments use a real trajectory dataset with pertinent weather observations to demonstrate the effectiveness and efficiency of the DICLERGE algorithm to the aircraft trajectory clustering problem.