Outlier Detection Approach for Discovering Anomalous Maritime Profiles

EasyChair Preprint 10468

Abstract

Maritime authorities play a key role in ensuring the safety and security of shipping lanes and ports. The port state control mechanism enables these authorities to physically verify suspect vessels (e.g., involved in smuggling or piracy events), but choosing the most relevant vessels to be inspected represents a challenging task. This decision can be enhanced by AI-powered systems that analyse large amounts of data, identify patterns and report all observed discrepancies. This paper presents a statistical analysis on the temporal durations of four types of naval statuses: sailing, docked in port, waiting at anchor and not transmitting AIS data. These durations were extracted from the historical activity of different classes of vessels that passed the Black Sea region (Romanian Exclusive Economic Zone) in 2022. Probability density functions were built for these vessels and all statuses’ durations were fitted into known parametric distributions. Finally, the paper shows the results of multiple outlier detection algorithms that searched for anomalous data in a multivariate manner.

Keyphrases: AIS, Port State Control, maritime anomaly detection, outlier detection, unsupervised learning

@booklet{EasyChair:10468,
  author    = {Alexandru Pohontu and Robert Gheorghe and Constantin Vertan},
  title     = {Outlier Detection Approach for Discovering Anomalous Maritime Profiles},
  howpublished = {EasyChair Preprint 10468},
  year      = {EasyChair, 2023}}