Detecting Abnormal Ship Trajectories Using Functional Isolation Forests and Dynamic Time Warping

This paper studies an algorithm allowing the iso-lation forest method to be adapted to time series associated with ship trajectories. This algorithm builds decision trees using different similarity measures between the ship trajectories of interest and the atoms of a dictionary constructed by the user. The similarity measure used to compare trajectories with potentially different lengths is based on dynamic time warping. Results obtained on synthetic data with an available ground truth yield promising results, when compared to the state-of-the-art.