Towards a Methodology for Data-Driven Automatic Analysis of Animal Behavioral Patterns

Measurement of behavior a major challenge in many animal-related disciplines, including ACI. This usually requires choosing specific parameters for measuring, related to the investigated hypothesis. Therefore, a key challenge is determining a priori what parameters are informational for a given experiment. The scope of this challenge is raised even further by the emerging computational approaches for animal detection and tracking, as automatizing behavioral measurement makes the possibilities for measuring behavioral parameters practically endless. This paper approaches these challenges by proposing a framework for guiding the decision making of researchers in their future data analysis. The framework is data-driven in the sense that it applies data mining techniques for obtaining insights from experimental data for guiding the choice of certain behavioral parameters. Here, we demonstrate the approach using a concrete example of clustering-based analysis of trajectories which can identify 'prevalent areas of stay' of the animal subjects in the experimental setting.