Microcontroller-Based EdgeML: Health Monitoring for Stress and Sleep via HRV
The healthcare sector is undergoing a transformation with the integration of cutting-edge technologies such as machine learning (ML), the Internet-of-Things (IoT), and Cyber–Physical Systems (CPS). However, traditional ML systems often face challenges in real-time processing and resource efficiency,...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2024-12-01
|
| Series: | Engineering Proceedings |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2673-4591/78/1/3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The healthcare sector is undergoing a transformation with the integration of cutting-edge technologies such as machine learning (ML), the Internet-of-Things (IoT), and Cyber–Physical Systems (CPS). However, traditional ML systems often face challenges in real-time processing and resource efficiency, limiting their application in life-critical scenarios. This research explores the potential of edge ML, particularly TinyML with TensorFlow Lite, implemented on microcontroller-based AI sensors for real-time health monitoring. By leveraging model quantization, the system analyzes heart rate variability (HRV) data to deliver continuous and personalized insights into stress levels and sleep quality. Trained on SWELL and ISRUC datasets, the system is highly energy-efficient, consuming 33 mW in idle mode, 66 mW during data collection, and 99 mW during real-time inference, making it suitable for resource-constrained environments. Performance analysis reveals significant demographic variations: younger individuals (18–25) achieved 90% accuracy due to higher HRV and lower baseline stress, while middle-aged (26–50) and older adults (50+) demonstrated declining HRV, reducing accuracy to 82% for the latter. Gender differences were also observed, with males exhibiting greater stress response sensitivity and better accuracy (89%) compared to females. This study underscores the transformative potential of TinyML for real-time, energy-efficient health monitoring and emphasizes the need for demographic-specific optimizations to enhance system reliability and accessibility. |
|---|---|
| ISSN: | 2673-4591 |