Anomaly Detection in Maritime Domain based on Spatio-Temporal Analysis of AIS Data Using Graph Neural Networks

Abstract

One of the critical applications of trajectory mining is Event Detection, where an automatized task identifies the deviation of a vessel’s movement from its standard route. Conventionally, Origin-Destination matrix data is utilized for event detection which has limitations such as removal of temporal aspect of data and inability to access to trajectory features such as speed of vessel from origin to destination. To utilize aforementioned features available in Automatic information system (AIS), we formulate the problem in a novel way, by detecting anomalies in a set of directed graphs representing the movement pattern at each time interval. We further propose Graph Network Deviation Detector (GNDD), which leverages graph embedding and context embedding techniques to capture anomalies in the spatio-temporal patterns of movement. Extensive experiments applied on five real-world AIS datasets show that our method achieved promising results in identifying abnormal movements.