Machine Learning-Based Species and Breeding Stage Classification in Animal Movement Trajectories

Oceans are environments where a diversity of human activities threaten marine life. To achieve effective conservation, it is crucial to comprehend the movement patterns of animals within these dynamic environments: how, when, where, and why they move. Until recently, data-driven analysis has been limited by data availability. In this talk, we will discuss the potential of novel machine learning techniques with the help of an extensive dataset encompassing 13,000 individual trajectories from over 100 marine species. Our primary focus will center on classification algorithms, specifically targeting species and breeding stage identification. We will scrutinize the factors influencing the performance of these algorithms and explore the realm of transfer learning for breeding stage classification. This revolutionary approach leverages knowledge acquired from one species to increase the accuracy in other species the model has not been trained on (zero-shot learning).