From Protection to Prediction: Exploring Wearables and AI for Smarter US Health Outcomes

Abstract

The growing intersection between wearable tech and machine learning is starting to reshape how we think about healthcare, not as something that reacts to illness after it appears, but as a system that can anticipate problems and help prevent them before they escalate. In this study, we explored how data from both consumer and clinical-grade wearables could be used to predict early signs of chronic conditions like heart disease, sleep disorders, and mental health issues in the U.S. We built a modeling framework that combined classic supervised algorithms, Random Forest, XGBoost, and logistic regression, with deep learning architectures designed to handle time-series data, including LSTMs and Transformers. The input came from a range of biometric signals: heart rate variability, skin temperature, motion, and sleep patterns. These weren’t toy datasets either, we trained and validated on real-world data pulled from healthcare providers, device manufacturers, and open health repositories. What we found was encouraging. LSTM models were especially good at picking up on temporal signals and subtle shifts over time, beating out the traditional models in most tasks. On average, they achieved a 93.1% AUC across our main prediction goals. One of the most noticeable gains came from combining wearable data with existing health records: doing so improved early detection accuracy by around 22% compared to using electronic health records alone. We also experimented with semi-supervised learning to tackle mental health classification, using unlabeled wearable data to help detect emotional state changes. That approach showed real promise, especially in identifying people who might benefit from early outreach or intervention. More broadly, the study points to how wearable data, when interpreted through the right models, can support real-time, always-on health monitoring. It’s not a replacement for traditional clinical care, but it could be a powerful layer that runs alongside it, quietly tracking, learning, and flagging issues before they become crises. Of course, making this work at scale means dealing with some non-trivial challenges around data governance, privacy, and interoperability. But if those foundations are built right, there's a clear path toward public health systems that are not only more responsive but more preventive by design.