HCAR: Hygiene and Cleanliness Activity Recognition using Smartwatch

Abstract

Maintaining personal hygiene and environmental cleanliness is critical for health, but daily monitoring of these behaviors remains challenging. In this study, we proposed HCAR (Hygiene and Cleanliness Activity Recognition), an on-device system for real-time recognition of personal hygiene and cleanliness activities using smartwatches. While our system achieved high accuracy (96.30%) in controlled environments, its performance significantly decreased in real-world conditions due to environmental noise and variability in activity. To address this challenge, we applied a personalization approach, by fine-tuning the model using user-specific data. Results demonstrated that 420 seconds of personal data per activity significantly restored the system's accuracy, achieving a balanced accuracy of 90.75%, which is suitable for practical daily monitoring. This study provides practical guidelines on the minimal amount of user data needed for effective personalization and lays the groundwork for future research in personal hygiene monitoring and activity interventions.