Abstract

Ethology, the study of animal behavior, seeks to understand the minds of animals. With recent advancements in technology, especially the rise of artificial intelligence (AI), researchers now have powerful new tools at their disposal. AI-based video analysis makes it possible to collect data automatically and non-invasively, without sensors attached to the animals, allowing scientists to handle larger datasets with less manual work and fewer errors.

In this project, open-source AI algorithms were used to study paradise fish, a species known for its complex behaviors. The algorithms were trained to do two main tasks: describe what the fish was doing (classification) and track its body movements over time (quantification). To set this up, we first recorded videos of the fish, identified the key body points we wanted to track across the frames, and labeled these points on carefully selected video frames. The algorithm was then trained on this labeled data and tested to ensure that no overfitting occurred and overall precision was adequate before the final analysis.

We focused on a specific behavior called surface respiration, where the fish comes to the surface to breathe air. The goal was to determine if the fish arches its back during this action. The results suggest it does, likely due to anatomical implications.

This study shows how AI-driven approaches can reveal details of animal behavior that are hard to detect through traditional methods, demonstrating its potential to deepen our understanding of why animals act a certain way.