Evaluation and comparison of anomaly detection algorithms in annotated datasets from the maritime domain

Anomaly detection supports human decision makers in their surveillance tasks to ensure security. To gain the trust of the operator, it is important to develop a robust system, which gives the operator enough insight to take a rational choice about future steps. In this work, the maritime domain is investigated. Here, anomalies occur in trajectory data. Hence, a normal model for the trajectories has to be estimated. Despite the goal of anomaly detection in real life operations, until today, mostly simulated anomalies have been evaluated to measure the performance of different algorithms. Therefore, an annotation tool is developed to provide a ground truth on a non-simulative dataset. The annotated data is used to compare different algorithms with each other. For the given dataset, first experiments are conducted with the Gaussian Mixture Model (GMM) and the Kernel Density Estimator (KDE). For the evaluation of the algorithms, precision, recall, and f1-score are compared.