Researchers from Apple, in collaboration with the University of Southern California, have developed a new artificial intelligence (AI) model that tracks the behavior data compared to sensor signals. The new research builds the Apple Heart and Movement Study (AHMS) in advance work, and the purpose was to understand that if behaviors, such as sleep sample and phase counting, traditional indicators such as heart rate and blood oxygen levels can be better decisive than a person’s health. According to the paper, the AI model performed amazingly, even if there are some warnings.
Apple’s new study shows the benefits of moving beyond traditional health data
The study titled “Beyond Censor Statistics: Wear Foundation Foundation Models Improve Health Predictions” was published in the Print Print Journal Archives and has not been reviewed. Researchers have come out to develop an AI model, which has been named wearing a wearing treatment model (WBM), which relies on a wearing viable data, such as how long a person sleeps and his REM cycle, daily steps and moves, and how to change his activity during the week.
Traditionally, to predict or evaluate one’s health, capable wear health research has generally focused on reading RAW sensor such as permanent monitoring of heart rate, oxygen levels in the blood and physical temperature. Studies believe that although these figures may be useful at times, it also lacks complete context about the individual and may have contradictions.
Regardless of, so far, the behavior data, which is also the most viable to wear, has not been used as a reliable indicator of a person’s health in the system. According to the study, there are two main reasons. First of all, these figures are much wider than sensor data, and as a result, it can be a lot of noise. Second, creating the algorithms and systems that can collect and analyze this data and make health predictions reliable.
This is the place where a large model of language (LLM) comes and solves the problem of analysis. To resolve the noise in the data, the researchers fed the model with structural and processed data. The data itself is derived from more than 1,62,000 Apple Watch users who participated in AHMS research, which has a total of 2.5 billion hours of wear data.
Once trained, the AI model used 27 different behavior measurements, which were added to the category such as activity, cardiovascular health, sleep and movement. It was then tested in 57 different precision tasks, such as finding out whether someone had a special medical condition (diabetes or heart disease) and tracking temporary health changes (retrieval from injury or infection). Compared to baseline accuracy, researchers claimed that WMB performed in 39 out of 47 results.
Comparison test model and both of the WBM model performance
Photo Credit: Apple
The results from the model were then compared to another test model, which was only fed to the raw heart data, also known as the Photo Plotesmogram (PPG) data. Interestingly, when compared individually, there was no clear winner. However, when researchers mixed both models, the prediction and health analysis accuracy was more understood.
Researchers believe that connecting traditional sensor data with behavior data can improve accuracy in the prediction of the health situation. The study states that the interpretation of the measurement data is easy, is better align with real -life health results, and is less affected by technical errors.
In particular, this study also highlighted several important boundaries. This data was taken from Apple Watch users in the United States, and did not represent the wider global population. In addition, due to the high cost of wearing equipment that accurately deposit and store the behavior data, rescue health care is also a challenge.
