Amending the European fishing fleet segmentation based on machine learning and multivariate statistics

Considering the critical issue of overexploited stocks due to overfishing, the EU’s Data Collection Framework (DCF) was established. Within the DCF, member states collect and analyze data relevant to sustainable fisheries management. To evaluate the status of fisheries, it is necessary to categorize fishing fleets into fleet segments. However, the current DCF segmentation is primarily based on technical vessel parameters, such as vessel length and predominant fishing gear, which often do not accurately represent the fishing activities of the vessels. To address this, we developed an alternative fleet segmentation approach that provides a more realistic overview of fishing activities. This approach utilizes multivariate statistics and is coupled with machine learning techniques for automatization. Applying this approach to two decades of German fisheries data resulted in a data set with fewer segments compared to the DCF approach, which represented the actual fishing strategies more closely. The comparison of biological stock health indicators calculated for both the current and the novel segmentation schemes revealed that the current scheme often misses signs of segments relying on overexploited stocks. The machine learning technique applied showed high classification accuracy, with misclassifications being rare and only occurring in segments with overlapping catch composition. Since machine learning enables almost perfect allocation to the revised segments, we expect a successful implementation of this protocol for future fleet segmentation. This approach is highly suitable for data collection and analysis procedures and can serve as a standard tool. Therefore, this novel approach can contribute to the improvement of fishing fleet analyses and policy advice for better fisheries management.