Apple Watch can already give you a lot of information on your body, such as your heart rate or your physical activity. However, researchers have found a new approach to better predict your state of health, using an AI that analyzes your behavior.
Connected watches are full of sensorssensorswhose potential is still far from fully exploited. In a study that will be presented at the ICML 2025 conference in Vancouver this weekend, researchers from AppleApple used artificial intelligence to predict the health of the carriers of the Apple Watch, not directly with the sensors, but from their behavior.
Rather than analyzing the data in real time, such as the heart rate, the researchers used AI to study the information calculated by the watch, such as the durationduration of sleepsleep or the number of steps. These are all 27 indicators provided by Healthkit. For this, the researchers were able to use the data of 161,855 participants, collected for 2.5 billion hours through theApple Heart and Movement Study.
A new approach to connected health
Scientists have thus created a new type of AI model that they called WBM (Wearable Behavior Model), which has proven to be more effective than previous models (PPG) trained for the analysis of raw data from the sensors in most situations. WBM notably had better scores for the prediction of what they call static health states, such as taking Beta-blockersBeta-blockersor transitional states, such as pregnancypregnancy. If its performance is better for most health states (39 out of 47), this is not always the case, such as for the diabetesdiabetes.
The researchers then combined the two types of AI to create a model hybridhybrid. The results were almost systematically better than with each of the models taken separately. According to the researchers, these results would allow early detection and more proactive and personalized care. It is therefore not impossible that Apple decides to optimize this model to integrate it into a next version of the Apple Watch, or even offer it via an update of theapplicationapplication Health.
