AccelerATE: Accelerometer Based Quantification of Dog Chewing

Chewing on bones in the domestic canine is an ancient and instinctive behavior. Recently there has been emerging interest by canine psychology researchers to explore possible correlations between chewing and stress reduction. The long-term study of dog chewing necessitates technology which can detect and quantify bites as dogs chew on toy bones. This technology could potentially be applied in home environments to assess and monitor stress in pet dogs. Previous experiments have employed environmental or body-worn microphones to capture audio from dogs chewing in controlled laboratory experiments, but this approach creates challenges in a home environment. In addition to ambient noise interfering with chewing sounds, the presence of a microphone in the home creates privacy concerns as the microphones could record private sounds such as conversations. To mitigate this, we hypothesize that bites can be detected and measured by a collar-worn accelerometer with similar accuracy to microphones. For this proof-of-concept study, we developed a prototype collar-worn accelerometer-based device and collected data with five dogs chewing on various chew toys. We conducted two machine learning experiments with a random forest classifier and achieved 67% and 70% accuracies averaged across a 5fold dog-independent cross-validation and a 10-fold dog-dependent cross-validation respectively. We believe further data is necessary in order to determine the viability of collar-worn accelerometers to detect chewing.