Global Fishing Watch processes data using machine learning to identify where and when fishing is occurring on a global scale. We’re employing a class of machine learning models called neural networks (NNs) to help us process all those AIS or VMS messages and to help us determine commercial fishing activity based on the vessel’s movements on the water. To determine when vessels are fishing, we’re using a NN to assign a “yes” score when vessels are actively fishing and a “no” score when they aren’t. We also designed a NN to figure out what types of vessels we are seeing—whether they are longliners, purse seiners, trawlers, or other.
In both cases, we give the computer a huge amount of AIS data that we’ve already analyzed and labeled by fishing score or vessel classification depending on what we’re training it to find. We call that the training data set. The computer sifts through it, finding patterns and determining which features are relevant and which aren’t. The computer then creates an algorithm, a set of rules that can be used to evaluate other AIS data and tell us what we want to know.