Quantified Canine: Inferring Dog Personality From Wearables

TL;DR

This study introduces a wearable device called 'Patchkeeper' that uses activity data to predict dog personality traits, enabling cost-effective and scalable personality assessment in real-world settings.

Contribution

We developed and validated a wearable sensor system that accurately predicts dog personality traits from activity data using machine learning, reducing reliance on expert assessments.

Findings

Achieved AUCs between 0.63 and 0.90 in predicting dog personality.

Collected 1300 hours of activity data from 12 dogs in natural environments.

Demonstrated the feasibility of using wearables for psychological assessment of pets.

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.