[ ] Come up with a clustering (unsupervised learning) machine learning algorithm to cluster vessels by behaviour instead of having set categories, and check how that overlaps with our current categories
[ ] Treat the output of the classifier as a likelihood of a vessel being of a certain gear type, where it can simultaneously be of multiple gear types, the likelihoods adding up to 1. Then run the fishing/non-fishing classifiers for all the gear types, and use a weighted average of their scores (multiplying their "is fishing" score by the likelihood that the vessel is of that gear type).