Apple’s latest breakthrough in Artificial Intelligence (AI) comes in the form of a model that can detect pregnancy with 92% accuracy by analyzing behavioural data from iPhones and Apple Watches. The findings are detailed in a study titled “Beyond Sensor Data: Foundation Models of Behavioural Data from Wearables Improve Health Predictions.” This study introduces the Wearable Behaviour Model (WBM), which relies on higher-level health metrics—like sleep quality, mobility, and heart rate variability—rather than just raw sensor data, which can often be noisy and hard to interpret.
Training the WBM: How Apple’s AI Model Learns to Detect Pregnancy with Precision
The WBM was trained using over 2.5 billion hours of wearable data and outperformed older models that relied on low-level sensor inputs. Researchers built a pregnancy dataset from 430 pregnancies, using Apple Health, HealthKit, and photoplethysmography (PPG) data. They labelled the nine months leading to childbirth and one month after as “positive” weeks—representing active pregnancy or postpartum recovery—while other weeks were marked “negative.”
In addition to the pregnancy group, data from over 24,000 non-pregnant women under 50 were analyzed to improve model accuracy. This diverse dataset helped WBM distinguish between physiological signals linked to pregnancy and other normal variations in women’s health.
Smarter Health Insights with WBM to Detect Pregnancy with Precision
One of the key advancements of WBM is its use of expert-designed algorithms to convert raw sensor readings into meaningful behavioural metrics. These metrics are not only clearer but also better aligned with real health states. According to 9to5macthe model allows for more precise tracking and prediction of health changes over time, marking a significant step forward in wearable-based health monitoring and AI-powered diagnostics.
Summary:
Apple’s new AI model, the Wearable Behaviour Model (WBM), detects pregnancy with 92% accuracy using behavioural data from iPhones and Apple Watches. Trained on 2.5 billion hours of wearable data, it uses high-level health metrics for precise predictions, marking a major advancement in AI-driven health monitoring.
