Apple Watch’s New AI Predicts Your Health With 92% Accuracy

When you think of health tracking on your Apple Watch, you probably imagine heart rate graphs, blood oxygen levels, or ECG results. But a new Apple-supported study suggests your everyday behavior, like how you move, sleep, and exercise, could actually reveal more about your health than these biometric readings.

This finding comes from a preprint paper titled “Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions“. 

Researchers from the Apple Heart and Movement Study (AHMS) have developed a new AI foundation model called WBM (Wearable Behavior Model). And here’s why it’s groundbreaking: instead of focusing on raw sensor data like heart rate signals, WBM learns directly from your behavioral metrics, things like step count, walking pace, sleep duration, mobility, and VO₂ max. All the data your Apple Watch already tracks daily.

Why Behavioral Data Matters More

You might wonder, if the Apple Watch already collects heart rate and other sensor data, what’s the point of this model?

Well, sensor data shows what’s happening in a few seconds, like your current heart rate. But health conditions such as pregnancy, hypertension, or the effects of long-term smoking are detected over days, weeks, or months. That’s where behavioral data shines.

Unlike raw sensor data, behavioral metrics are more stable and directly linked to real-world health changes. For example, the study explains that changes in walking gait, overall activity levels, or sleep quality can help detect pregnancy or even early signs of many health issues.

What Makes WBM Special

The WBM was trained using data from over 161,000 participants, covering more than 2.5 billion hours of wearable data. But instead of feeding it raw streams, researchers used 27 human-readable metrics, including active energy burned, gait stability, respiratory rate, heart rate variability, and sleep duration.

They broke down this data into weekly chunks and used a new AI architecture called Mamba-2, which performs better for this kind of behavioral analysis than standard Transformer models like GPT.

When tested on 57 health tasks, WBM outperformed existing models that only used heart rate sensor data (PPG) in most cases. For static health states (like detecting if someone uses beta blockers), WBM performed better in 18 out of 47 tasks. For dynamic states (like pregnancy or sleep quality), it outperformed PPG in almost all cases, except diabetes, where raw heart rate data still worked better.

The Power of Combining Both Worlds

Interestingly, combining WBM’s behavioral data with PPG’s sensor data delivered the best results overall. This hybrid model achieved 92% accuracy in pregnancy detection and performed better in detecting sleep issues, infections, injuries, and heart conditions like Afib.

In short, your behavior + sensor readings = the most accurate health predictions.

What’s Next for Apple Watch?

This doesn’t mean your Apple Watch will stop showing heart rate graphs. Instead, Apple could integrate WBM alongside existing PPG or ECG sensors to enable even smarter health detection features. The company has always focused on making health tracking meaningful, and this study shows that behavioral AI could play a huge role in future Apple Watch health updates.

Imagine your watch telling you more about pregnancy risk, sleep quality, or heart health, not just based on your heartbeat but on how you live your everyday life. Because sometimes, your behavior speaks louder than your heart.

Don’t miss these related reads:

• The history of the Apple Watch: 10 years and counting…
• Everything we know about the upcoming Apple Watch Series 11
• Apple Watch Ultra 3 Latest Leaks: Features, Specs, & Release Date