A ML-Based Multi-Species Analysis to Explore the Environmental Drivers of Cetaceans Behaviors

This study investigated environmental drivers of behavior activity states in three odontocetes species (Risso's dolphin, common bottlenose dolphin and striped dolphin) inhabiting the Northern Ionian Sea. By analyzing their behavioral patterns in relation to 20 environmental variables derived from datasets such as the Copernicus Marine Service, Machine Learning (ML) models employing either Random Forest (RF) or RUSBoost algorithms were developed for multi-species behavioral classification. Key findings included the RUSBoost model achieving 94% accuracy, 93% sensitivity, and 100% specificity in predicting feeding behavior between Risso's dolphin and common bottlenose dolphin. Additionally, RF model excelled in classifying socializing and traveling behaviors between Risso's and common bottlenose dolphin. By focusing on key environmental drivers and their influence on different species' behaviors, this study proposed a robust ML-based methodology for behavioral analysis, showing potential for applications to other behaviors and species, and also highlighting its investigative capacity for broader ecological studies.