Understanding how animals behave and move is important to improve wildlife monitoring and management, especially for critically endangered species such as the California condor. Sophisticated tracking devices called accelerometers are providing new insights into behaviors such as hunting, foraging, and migration based on acceleration. However, equally sophisticated methods for interpreting these data are needed. Researchers compared the effectiveness of unsupervised and supervised machine learning tools for analyzing California condor accelerometry data. Unsupervised methods look for similarities within the movement data and divide them according to natural breaks in the data set. Supervised machine learning methods are trained to recognize specific behaviors through examples, then automatically assign data to categories. The team found that unsupervised methods of classification of accelerometry data are not as suitable as supervised techniques for identifying specific behavior types. The ability to remotely monitor endangered species and efficiently identify when and where certain behaviors are happening provides researchers and managers with a powerful tool for wildlife conservation.  

Sur, M., Hall, J.C., Brandt, J., Astell, M., Poessel, S.A., and Katzner, T.E., 2023, Supervised versus unsupervised approaches to classification of accelerometry data: Ecology and Evolution, v. 13, no. e10035, Online. https://doi.org/10.1002/ece3.10035