Real-time health monitoring for elderly people at home using AI and wearable devices

What is it about?

Many health problems do not appear suddenly but develop gradually through small changes in the body or daily behavior. However, most healthcare systems only check patients occasionally, which means early warning signs are often missed. In this work, we present AI on the Pulse, an artificial intelligence system that continuously monitors people using everyday wearable devices (such as smartwatches) together with sensors placed in the home environment. The system learns what is normal for each individual by observing their heart rate, heart rate variability, breathing, sleep, physical activity, and environmental conditions like temperature or air quality. Instead of trying to recognize specific diseases, the system looks for unusual changes compared to a person’s own baseline. These changes, called anomalies, can indicate stress, cardiovascular issues, sleep problems, or other health risks. When an anomaly is detected, the system sends a real-time alert to healthcare professionals. To make these alerts easier to understand, large language models automatically generate clear, medically meaningful explanations that describe what changed and why it may matter. The system has been tested both on public medical datasets and in a real home-care study with older adults using only non-invasive, consumer-grade devices. The results show that reliable and continuous health monitoring is possible without hospital equipment.

Featured Image

Photo by Káplár Bálint Áron on Unsplash

Photo by Káplár Bálint Áron on Unsplash

Why is it important?

This work addresses a key limitation of current digital healthcare: the lack of continuous, personalized monitoring in real-world settings. Unlike many existing systems that require labeled medical data or hospital-grade sensors, our approach works without constant manual labeling and adapts to each patient’s individual physiology and lifestyle. Technically, the system improves the state of the art in time-series anomaly detection, achieving substantially higher accuracy than existing methods across multiple datasets. Practically, it demonstrates that early health warning signals can be detected at home, using lightweight and affordable technology. Equally important is the focus on interpretability. By translating complex AI outputs into clear clinical explanations, the system supports trust and usability for healthcare professionals. This makes the approach suitable for real deployment in home care, chronic disease management, and aging populations, where early intervention can improve outcomes and reduce healthcare costs.

Perspectives