Scholarlink

In the early 2000s, AIS (Automatic Information System) became mandatory for almost all seagoing vessels, which must broadcast the ship’s tracking and navigational information for collision avoidance and maritime safety. Maritime surveillance agencies can use AIS data for the classification of vessels, e.g., for the detection of spoofing behavior in fishery scenarios. This contribution looks at detecting Fishery type behavior by Random Forests based on aggregated AIS messages. It addresses the question of how much classification performance, measured by accuracy and Cohen’s kappa, depends on the geographical features in training datasets of AIS data rather than on general fishery vessels’ properties. For this purpose, Random Forests are trained on aggregated real-life AIS data for fishery detection in different maritime areas and are then group-cross-validated against each other.

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

Cross-Validation of Random-Forests’ Classification Performance in Maritime Scenarios with Aggregated AIS Messages

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Max Krüger

摘要

In the early 2000s, AIS (Automatic Information System) became mandatory for almost all seagoing vessels, which must broadcast the ship’s tracking and navigational information for collision avoidance and maritime safety. Maritime surveillance agencies can use AIS data for the classification of vessels, e.g., for the detection of spoofing behavior in fishery scenarios. This contribution looks at detecting Fishery type behavior by Random Forests based on aggregated AIS messages. It addresses the question of how much classification performance, measured by accuracy and Cohen’s kappa, depends on the geographical features in training datasets of AIS data rather than on general fishery vessels’ properties. For this purpose, Random Forests are trained on aggregated real-life AIS data for fishery detection in different maritime areas and are then group-cross-validated against each other.