Challenges in Vessel Behavior and Anomaly Detection: From Classical Machine Learning to Deep Learning

Abstract

The global expansion of maritime activities and the development of the Automatic Identification System (AIS) have driven the advances in maritime monitoring systems in the last decade. Given the enormous volume of vessel data continuously being generated, real-time analysis of vessel behaviors is only possible because of decision support systems provided with event and anomaly detection methods. However, current works on vessel event detection are ad-hoc methods able to handle only a single or a few predefined types of vessel behavior. Most of the existing approaches do not learn from the data and require the definition of queries and rules for describing each behavior. In this paper, we discuss challenges and opportunities in classical machine learning and deep learning for vessel event and anomaly detection.