Methods for discovering models of behavior: A case study with wild Atlantic spotted dolphins

Abstract

We present an interactive data mining method to support marine mammalogists investigating communicative behaviors with wild Atlantic spotted dolphins, Stenella frontalis. Most current tools require scientists to inspect behavior data manually, form hypotheses, create a detector for patterns of interest, search their database of behavior, and measure the results to determine if their hypothesis was correct. In contrast, our technique discovers patterns in the data automatically, allowing researchers to associate these patterns with behaviors of interest. By visualizing where in time these patterns occur and in what orders, the scientist can quickly form hypotheses and then use statistical tests on a reserved test set of data to determine whether those hypotheses have support. Our hope is that these methods will speed behavioral research significantly.