Quantified Canine: Inferring Dog Personality From Wearables

Lakmal Meegahapola,Marios Constantinides,Zoran Radivojevic,Hongwei Li,Daniele Quercia,Michael S. Eggleston
DOI: https://doi.org/10.48550/arXiv.2301.06964
IF: 6.4588
2023-01-17
Human-Computer Interaction
Abstract:Being able to assess dog personality can be used to, for example, match shelter dogs with future owners, and personalize dog activities. Such an assessment typically relies on experts or psychological scales administered to dog owners, both of which are costly. To tackle that challenge, we built a device called "Patchkeeper" that can be strapped on the pet's chest and measures activity through an accelerometer and a gyroscope. In an in-the-wild deployment involving 12 healthy dogs, we collected 1300 hours of sensor activity data and dog personality test results from two validated questionnaires. By matching these two datasets, we trained ten machine-learning classifiers that predicted dog personality from activity data, achieving AUCs in [0.63-0.90], suggesting the value of tracking the psychological signals of pets using wearable technologies.
What problem does this paper attempt to address?