Predicting Depression and Anxiety Mood by Wrist-Worn Sleep Sensor

In recent years, researches on recognizing daily behavior and psychological/physiological states have been actively conducted to change the behavior of workers working in companies. In this paper, we analyzed Occupational Health questionnaire named DAMS for waking-up time and daily sleep data that are acquired from wearable devices in 2-3 weeks experiment of 60 office workers working in five general companies. By using a machine learning method, our binary Balanced Random Forest model predicts depression, positive, and anxiety moods in two levels, high and low. As a result of Leave One Person Out cross validation, it was confirmed that our model estimated with the F1 values of depression mood: 0.776, positive mood: 0.610, anxiety mood: 0.756. Moreover, we evaluated the variance of the three estimations among subjects by referencing the box chart. As a result, it was confirmed that there is variance in estimation accuracy for each subject.