A Machine Learning Framework for Circadian Stability Assessment Using Wearable Health Data

Abstract

Circadian rhythm affects multiple physiological and behavioral processes including sleep, metabolism, hormone secretion, and cognitive functions. A mismatch in circadian rhythm has been associated with sleep issues, decreased productivity, metabolic conditions, and even a broader set of health risks. The current paper introduces a machine learning solution for the evaluation of circadian rhythm stability based on physiological and behavioral data obtained from wearables. The solution is based on the combination of metrics related to sleep, heart rate variability (HRV), physical activity levels, chronotype characteristics, and questionnaire responses in order to compute the overall circadian stability score. A Random Forest Regressor is applied in order to capture the interactions between the factors mentioned above and to predict circadian stability with recommendations on lifestyle choices. According to the experimental results, the $\mathbf{R}^{\mathbf{2}}$ score is 0.87 which indicates high precision of predictions along with high interpretability of the model. The framework provides a number of insights that can help in preventive healthcare, healthier lifestyle choices, and continuous monitoring of circadian health status. Thanks to the lightweight and scalable design, the proposed solution can be implemented in wearable health monitoring solutions to assess the circadian rhythm in real-time.