Apple Watch data fed an AI – what it found might change everything

A new AI model trained on over 2.5 billion hours of Apple Watch data may predict health conditions better than traditional methods.

1comment
Apple Watch data fed an AI – what it found might change everything
A new research paper has been published on an AI model trained on Apple Watch data, which can now predict health conditions more accurately. Apparently, it's more accurate than traditional sensor-based approaches, according to a newly published study

The research paper is titled "Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions". The study shows a machine learning model that analyzes user behavior with the aim of flagging potential health issues. 

Earlier models of this kind were focused on sensor outputs (in real time) like heart rate or blood oxygen. However, this new model also identifies patterns in how you move, sleep, or exercise over time. 

A foundation model called the Wearable Behavior Model (WBM) sits at the center of the study. This model analyzes high-level behavioral metrics, including step count, sleep duration, variability of heart rate, and mobility. These are all calculated by the Apple Watch with on-device algorithms. 


Apparently, according to the researchers, this approach allows the AI model to detect certain health conditions more effectively than using only biometric data. The WBM identified the best the so-called static health states (when a person takes beta blockers, for example) or transient health conditions like sleep quality or respiratory infection. 

How do you feel about AI using your wearable data for health predictions?

Sounds amazing – anything that helps detect issues early
41.67%
I'm cautiously optimistic, but privacy is key
41.67%
Not a fan – feels like a step too far
0%
I'd need more proof it actually works
16.67%


The model also achieved 92% accuracy for pregnancy detection (combined with traditional biometric data in a hybrid approach). 

Apple collected data for this model through the Heart and Movement study.  In this study, more than 1600,000 participants voluntarily shared their data via the Apple Watch and their iPhones. The model was trained on over 2.5 billion hours of data (can you even imagine this, by the way?) and evaluated on 57 different health prediction tasks. 

Its algorithm identifies changes in behavior over time (days or weeks). This helps it identify conditions that happen over time, rather than sudden, acute ones. 

The researchers claim that smartwatches have now evolved to a point where they can support and offer data to this kind of AI-powered analysis. It's not clear whether such a model will be integrated into an Apple Watch in the future. However, it does show that current Apple Watch hardware is promising and could do much more when it comes to health analysis. 

Grab the Galaxy S25 + 2 Yrs Unlimited – only $30/mo from Mint Mobile

With Galaxy AI – port-in & $720 upfront required


We may earn a commission if you make a purchase

Check Out The Offer
Loading Comments...

Latest Discussions

by TBomb • 3
by bklabel • 15

Recommended Stories

FCC OKs Cingular\'s purchase of AT&T Wireless