As the most diverse group of vertebrates, fish exhibit swimming modes that critically determine their ecological adaptation and survival strategies. Different swimming modes vary in propulsive efficiency and maximum attainable speed (Webb, 1984). Since the seminal work on fish locomotion by Breder, (1926), understanding of fish movement patterns has advanced significantly. Current frameworks predominantly rely on the 12-category classification proposed by Vatanabe et al. (2008). However, traditional studies often assume single-mode swimming, neglecting multi-modal behavioral shifts across speed gradients. Here, we integrated FishBase data, original morphological trait datasets (17,496 entries), and genus-level swimming videos (1,673 genera) to establish a refined 25-category classification system. Using 42 morphological features and a CatBoost machine learning model (97.55% accuracy), we predicted swimming modes for 87.9% of modern fish genera, systematically mapping their global distribution patterns. Correlating results with the IUCN Red List revealed extinction risks associated with specific modes (e.g., Tetradondtiform), suggesting locomotor strategies influence environmental adaptability. This work advances fish movement classification and introduces a novel dimension for biodiversity conservation prioritization.
 

Fish lomocation,Actinopterygii,Machine learning,Catboost,Extinction risk