Scholarlink

The Automatic Identification System (AIS) data, which was initially conceived for collision avoidance, has increasingly been repurposed for surveillance applications. However, the intrinsic complexity and sheer volume of AIS data transmitted to base stations make humanly impossible to use it for surveillance. To mitigate this issue, we propose an unsupervised learning algorithm designed to autonomously identify anomalous instances within the AIS dataset throughout a vessel’s voyage. Given that the raw fields in AIS data are not directly applicable for training machine learning (ML) algorithms, we introduce an innovative feature extraction methodology to derive pertinent attributes essential for training the Isolation Forest. The trained Isolation Forest algorithm automatically detects the anomalous AIS instances automatically based on the extracted features. The number of instances which the officers need to be given attention is relatively very less compared to the actual AIS instances.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Isolation Forest-Based Anomaly Detection of Vessels Using AIS Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Supratim Manna,
  • R. Bharath

摘要

The Automatic Identification System (AIS) data, which was initially conceived for collision avoidance, has increasingly been repurposed for surveillance applications. However, the intrinsic complexity and sheer volume of AIS data transmitted to base stations make humanly impossible to use it for surveillance. To mitigate this issue, we propose an unsupervised learning algorithm designed to autonomously identify anomalous instances within the AIS dataset throughout a vessel’s voyage. Given that the raw fields in AIS data are not directly applicable for training machine learning (ML) algorithms, we introduce an innovative feature extraction methodology to derive pertinent attributes essential for training the Isolation Forest. The trained Isolation Forest algorithm automatically detects the anomalous AIS instances automatically based on the extracted features. The number of instances which the officers need to be given attention is relatively very less compared to the actual AIS instances.