In this project, we analyzed biodata from two biomonitoring devices, a Fitbit monitor worn on the wrist and a Basis monitor worn on both the ankle and the wrist (at different times).
The Fitbit monitor only measured steps per minute and calories burned per minute, but the Basis monitor measured steps per minute, heart rate, skin temperature, and galvanic skin response (also called electrodermal activity; this is a measure of nervous system activity, often used in EKGs and polygraphs).
Our primary interest was in analyzing the heart rate data, but the data contained many errors, especially when the monitor was worn on the ankle. We took two approaches to correcting these errors. First, we removed all Basis data points where the corresponding step data from the Fitbit differed by more than 50%. Second, we smoothed the raw Basis data by calculating a moving average over five minute intervals. Finally, we compared the statistical properties of the raw and corrected data sets, both when the Basis monitor was worn on the wrist, and when it was worn on the ankle.
Attached is a notebook with graphs of the comparisons between raw, smoothed, and corrected data.
Attachments: