Toward personalized sleep-wake prediction from actigraphy
- 0 Collaborators
Sleep patterns vary considerably among individuals. Existing algorithms and commercial devices of sleep quality assessment are pre-trained for all. We advocate a personalized sleep quality assessment with machine learning algorithms trained specific to individuals. ...learn more
Project status: Published/In Market
Overview / Usage
We argue that each person sleep differently and because of that, we shouldn't train a single machine learning algorithm based on data coming from everybody. Instead we developed machine learning algorithms that are person-specific, i.e., trained on individual data without including data from other individuals. We showed that not only did not the performance of our developed machine learning algorithms decrease as a result of excluding population-level data (by only including data from one individual), but they actually perform well and capture the individual's sleeping patterns effectively.
Methodology / Approach
We systematically developed 5 different families of personalized machine learning algorithms and trained them on individual-level data.
Technologies Used
Implementations were all done using Python's Sklearn library.