A comparison of supervised learning schemes for the detection of search and rescue (SAR) vessel patterns

Abstract

The overall aim of this work is to perform a systematic analysis of several off-the-shelf machine learning classification algorithms and to assess their ability to classify Search And Rescue (SAR) patterns from noisy Automatic Identification System (AIS) data. Specifically, we evaluate Decision Trees, Random Forests and Gradient Boosted Trees on a large volume of historical AIS data so as to detect SAR activity from vessel trajectories, in a scalable, data-driven supervised way, with no reliance on external sources of information (e.g. coast guard reports). Our analysis verifies that it is possible to identify SAR patterns, while the results show that although all algorithms are capable of achieving high accuracy, Random Forests marginally outperform the others in terms of performance and speed of execution.