Estimation of fish feeding groups, dietary breadth and trophic levels based on open data through the integration of TroPhish and FishBase data

TroPhish investigated how well freshwater fish feeding data could be integrated with Fish-base's species, ecology, diet, and nutrient component tables. It also explored how open datasets downloaded from Fish-base could be used to model fish feeding groups, feeding diversity, and trophic levels. For the analyses, 54,750 raw feeding records and 4,571 sample-level specified "fish ID" units were used as the basis. Fish-base matching was successfully completed for 88.6% of samples and 84.2% of species. This research led to the addition of "SpecCode" and "FamCode" to 4,051 samples, feeding-based trophic levels to 3,280 samples, and nutrient data to 3,440 samples for analysis. In the "Species Group K-Fold" validation, the Random Forest model was applied and enhanced with Fish-base datasets. In the feeding group classification, the trophic identification scenario achieved 0.515% accuracy, 0.397% balanced accuracy, and a 0.390% macro F1 score. In the training data, the model's generalizability to uncharacterized families was poor. The low performance in the "Family Code Group K-Fold" validation of the same scenario is evidenced by this. While the relational power of the feeding breadth regression was found to be limited (maximum R² = 0.092), the trophic level regression based on fish-based data yielded realistic findings in the taxon-trait scenario (R² = 0.403). The results suggest that the misleading signal related to fish feeding habitat is amplified by the inclusion of TroPhish and fish- based data. Model outputs are not an automated feeding classification system that produces definitive results. They should be considered as descriptive findings generated from research-driven modeling and publicly available data.