High-frequency accelerometer recording of key predatory behaviors in vipers: validation and case study with Timber Rattlesnakes (Crotalus horridus)

Morgan Thompson, Georgia College & State University


Tri-axial accelerometers (ACTs) are increasingly common in studies of animal behavior wherein direct observation of subjects in nature is limited. ACTs are small (< 1 g) piezo-electric sensors that measure three-dimensional acceleration derived from subject motion. When leveraged with advanced machine learning techniques, these data enable precise automated classification of a wide range of movement-mediated behaviors. Until recently, ACTs were once reserved for larger-bodied organisms most amenable to the temporary external attachment of devices. Ongoing ACT miniaturization has now expanded the breadth of organisms amenable to these methods. This project aimed to expand on a recently developed ACT ethogram developed for pit vipers by conducting intensive captive validation trials stimulating key foraging behaviors in Timber Rattlesnakes (Crotalus horridus). We completed 18 successful captive foraging trials, and these data were used to train and test three algorithms for automated classification of targeted behaviors (moving, still, strike, and swallow). The best performing model (Random Forest) correctly predicted strikes at 100% and swallowing at 90.83%. The high classification performance, even given a relatively limited training dataset, demonstrates the promise of this method for remote and continuous field monitoring of predatory behavior in pit vipers, a previously impossible task with conventional techniques, alone.