A Combined Approach to Predicting Rest in Dogs Using Accelerometers

The ability to objectively measure episodes of rest has clear application for assessing health and well-being. Accelerometers afford a sensitive platform for doing so and have demonstrated their use in many human-based trials and interventions. Current state of the art methods for predicting sleep from accelerometer signals are either based on posture or low movement. While both have proven to be sensitive in humans, the methods do not directly transfer well to dogs, possibly because dogs are commonly alert but physically inactive when recumbent. In this paper, we combine a previously validated low-movement algorithm developed for humans and a posture-based algorithm developed for dogs. The hybrid approach was tested on 12 healthy dogs of varying breeds and sizes in their homes. The approach predicted state of rest with a mean accuracy of 0.86 (SD = 0.08). Furthermore, when a dog was in a resting state, the method was able to distinguish between head up and head down posture with a mean accuracy of 0.90 (SD = 0.08). This approach can be applied in a variety of contexts to assess how factors, such as changes in housing conditions or medication, may influence a dog's resting patterns.