Identifying suspicious fishing activity based on AIS disabling events
In: Research Square (Preprints). Research Square: Durham. ISSN 2693-5015, more
|  |
| Available in | Authors |
|
Document type: Preprint
|
| Keyword |
|
| Author keywords |
fishing activity, AIS, fishing watch, illegal, unreported, unregulated, class imbalance, AIS disabling, cost-sensitive learning, oversampling, undersampling, neural networks |
| Authors | | Top |
- Agarwal, A.
- Gala, J.
- Mantha, S.
|
|
|
| Abstract |
A staggering loss of US$10 billion to US$23 billion is incurred each year due to illegal, unreported, and unregulated (IUU) fishing activities along with a severe loss to biodiversity. The Automatic Identification System (AIS), is a tool used to track vessel activity and avoid collisions. It is now being used to detect IUU activities as well, but it has a major drawback as the AIS transponders could be disabled due to various reasons, either illegal or otherwise, hence reducing its effectiveness. According to Welch et al. (2022), more than 55,000 suspected intentional disabling events (> 4.9M hours) occurred between 2017 and 2019. Thus the need for much more sophisticated global surveillance has increased and algorithms to analyze such huge amounts of data are required. We present a machine learning solution based on historical data to detect vessels of interest using the AIS Disabling Events dataset obtained from the Global Fishing Watch combined with the Regional Fisheries Management Organizations (RFMOs) datasets containing details of vessels caught in IUU fishing activities previously within their respective regions. One of our best models is the XGBoost with cost-sensitive learning boasting a minority recall of 0.79 and a majority recall of 0.76. |
|
