Consideration of Action Rhythm from Wearable Device data and missing pattern

Abstract

In recent years, various sensor devices have been released, and we can easily collect biological information. One of the familiar sensor devices is a wearable device. There are more than 100 types of wearable devices, and many devices can acquire activity and sleep information etc. Wearable devices are also being used in clinical studies etc., and daily life data which was difficult to obtain in the past can be acquired, and the possibility of new evaluation has been expanded. However, while we can acquire the data on day-to-day, missing data may occur in various situations. Of course, during charging data is missing, but data will miss during bathing, cooking, or forgetting to wear depending on the person. It's important to make missing as little as possible in order to perform high-accuracy analysis, and when make use of wearable device, it's desirable to control as avoiding missing except during charging. In this study, machine learning was used to predict the behavior of users based on the missing patterns, rhythms, and cycle etc. in addition to the acquired data from wearable device, the reason of missing was determined. And based on the results was considered, the usefulness of wearable device. This study was conducted based on data collected by Fitbit Charge3.