Study of swordfish fishing dynamics in the eastern Mediterranean by means of machine-learning approaches

We analyzed fisheries data collected in 2000 and 2001 from the Greek swordfish fishing fleets operating in the eastern Mediterranean, by means of machine-learning approaches, in order to define differences in exploitation patterns and fishing strategies. Based on their total annual catch, fishing vessels have been classified in three groups: low, medium and high producers. Decision-tree analysis revealed that group membership could be successfully predicted from the total number of working days per year, the vessel length, the type of gear used, the hook size and the number of hooks per set. Using the data of 2001 as a test data set and assuming that only the average catch of the most productive group was known, total production estimates for that year showed very little difference (7.92%) from the true values. These findings indicate that simple sampling schemes focusing on the high producers may be adequate for the examined fisheries. They also provide evidence that such methodological approaches could be useful for the cross-checking of fisheries estimates obtained through various sampling schemes.