Maritime transportation remains the go-to method when it comes to trading in bulk, with international shipments being responsible for close to 90% of goods. This provides an ideal platform for smuggling illegal goods as well, which in-turn makes maritime security a matter of great importance to businesses and governments alike. To combat against nefarious actors at sea, we cannot simply make more boats and have more boat police prowling the oceans. Have you seen the ocean? It’s huge. We need to target smarter and let the power of Machine Learning perform its magic. When applied correctly, Machine Learning can save time, resources and money. One effective way to utilize Machine Learning in the maritime domain is to apply it to the vessel’s Automatic Identification System (AIS).
Automatic Identification System was developed years ago as a safety mechanism for vessels transiting the seas. It is the primary system used to track the global fleet and is updated regularly to show actual navigational locations. It also rapidly transmits a variety of ship information between vessels and shoreside stations such as the ship’s weight, location, speed, MMSI (Maritime Mobile Service Identity) number, type of ship, and various other characteristics about the vessel or the voyage. Since this system originally launched back in the early 1990’s, there is a lot of historical data we can apply to machine learning to assess risk, performance, safety and economic health.
Analyzing historical data can lead to the implementation of learning algorithms that determine deviations in maritime environments. These deviations can, in turn, lead to informed risk assessments. Mitigating risk becomes simpler with Machine Learning when vast amounts of historical data are available. This data helps build a normal operating profile for each vessel which is the first critical step to applying a machine learning application.
There are a few different machine learning methods that help organize the AIS data. These machine learning methods are specific to anomaly detection within large datasets. Those methods involve supervised and unsupervised learning and rely on the validity of the data. I know you’ve heard of “crap in”, “crap out”, which is a simple way to say the data needs to be verified or the output is going to be useless. The larger the data set of known, corroborated events or activities, the more reliable the data. I’ll spare everyone the mathematics behind these methods because they can be extensive, but those wanting a deeper knowledge of them are encouraged to check out a thesis written by Jacob Coleman on implementing these methods when it comes to analyzing AIS data (https://core.ac.uk/download/pdf/322935023.pdf).
Bad actors in maritime environments consistently tamper with their AIS data to conceal their true identities or movements while at sea. And why not since they’re the ones inputting the data into the AIS device. They will either switch off the AIS device on the ship or they can even go as far as to send out false AIS data. This is becoming more popular among criminals. It’s because this data can be so easily manipulated that makes these machine learning methods/AI so important when it comes to maritime security. As bad actors continue to fib about their positional data in a consistent way, the AI has new deviations or classifications to isolate. The machine learning methods can only take us so far. Companies will still have to hire an analyst or a data scientist to make sense of whatever deviations or anomalies are discovered. This job will become more and more important as digital data informs and influences more marketplaces, not only maritime trade.
