본문 바로가기
  • Home

A Semi-Supervised Learning-Based Pet Behavior Classification System Using Wearable Devices

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(3), pp.31~42
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : February 6, 2025
  • Accepted : March 7, 2025
  • Published : March 31, 2025

JunHyeok Go 1 Siung Kim 1 JeongHyeon Park 1 Nammee Moon 1

1호서대학교

Accredited

ABSTRACT

This study proposes a wearable device-based system for classifying pet behaviors. The system incorporates a preprocessing step that effectively removes inactive data from unlabeled datasets using the DeepSVDD (Deep Support Vector Data Description) algorithm, thereby enhancing training performance. Furthermore, the system applies the MPL (Meta Pseudo Labels) method and contrastive learning techniques to extract and utilize meaningful information from unlabeled data during training. Experimental results demonstrated the system's effectiveness, achieving 87% accuracy with a labeled data ratio of 20% and 91% accuracy with a labeled data ratio of 60%. This system is practical for monitoring the health status of pets or detecting abnormal behaviors at an early stage. By integrating wearable technology with semi-supervised learning techniques, it presents new possibilities in the field of pet behavior analysis.

Citation status

* References for papers published after 2023 are currently being built.

This paper was written with support from the National Research Foundation of Korea.