Comparison of machine learning methods to predict sleep quality from daytime activity and nightly bedroom environmental conditions

The objective of this study is to use measurements of activity and indoor air quality during sleep as features to predict nine sleep quality metrics gathered from wearable devices and ecological momentary assessments. We compare the predictive capabilities of three machine learning models - logistic regression (LR), K-nearest neighbor (KNN), and random forest (RF). Results using data from a field study with 20 participants collected over 2.5 months indicate that the LR and RF models predict device-measured sleep efficiency and self-reported restfulness and sleep time with adjusted F1-scores of 0.90, 0.92, and 0.80, respectfully. Our results show that typical machine learning methods have the power to give insights into how people might sleep based on easily measured parameters.