TY - JOUR
T1 - Wearables data integration
T2 - Data-driven modeling to adjust for differences in Jawbone and Fitbit estimations of steps, calories, and resting heart-rate
AU - Shah, Yash
AU - Dunn, Jocelyn
AU - Huebner, Erich
AU - Landry, Steven
N1 - Funding Information:
To Tristan Bassingthwaighte, Sheyna Gifford, Christiane Heinicke, Carmel Johnston, Andrzej Stewart, and Cyprien Verseux, thank you for your invaluable contributions to this research. This research would not have been possible without the support of NASA Human Research Program as this HI-SEAS mission was funded by NASA, NNX13AM78G ?Key Contributors to the Maintenance and Regulation of Team Function and Performance on Long Duration Exploration Missions.?
Publisher Copyright:
© 2017 Elsevier B.V.
PY - 2017/4/1
Y1 - 2017/4/1
N2 - Differences in data output from two leading devices in the consumer-grade wearables market have been examined, namely Jawbone UP4 and Fitbit Charge HR devices, by comparing measurements that were conducted while participants wore both devices in tandem. Aggregate daily totals of steps and calories were shown to be highly correlated between devices (0.82–0.93 correlation coefficient for steps and 0.71–0.85 for calories); however, at the hourly level, differences in data output are evident, especially during hours of vigorous activity. These differences lead to both under- and over-estimation of measures such as hourly step-counts. Heart rate measurement with Jawbone and Fitbit is shown to be significantly different even at the daily level (p-value < 0.00001), which could be due to hardware differences in sensor type and possibly due to unknown differences in proprietary algorithms. Models were trained to enable adjustment of data collected from one device to the equivalent value in terms of the other device's measurement. This approach to data integration is recommended for researchers who are comparing data from multiple wearable devices, for individual users who have switched from one device to another and could use this method to adjust their wearables data history to be comparable with the new device, or for users who are comparing data with a user who has another type of device, or for groups organizing fitness challenges and health initiatives that can track users by comparing diverse wearables data.
AB - Differences in data output from two leading devices in the consumer-grade wearables market have been examined, namely Jawbone UP4 and Fitbit Charge HR devices, by comparing measurements that were conducted while participants wore both devices in tandem. Aggregate daily totals of steps and calories were shown to be highly correlated between devices (0.82–0.93 correlation coefficient for steps and 0.71–0.85 for calories); however, at the hourly level, differences in data output are evident, especially during hours of vigorous activity. These differences lead to both under- and over-estimation of measures such as hourly step-counts. Heart rate measurement with Jawbone and Fitbit is shown to be significantly different even at the daily level (p-value < 0.00001), which could be due to hardware differences in sensor type and possibly due to unknown differences in proprietary algorithms. Models were trained to enable adjustment of data collected from one device to the equivalent value in terms of the other device's measurement. This approach to data integration is recommended for researchers who are comparing data from multiple wearable devices, for individual users who have switched from one device to another and could use this method to adjust their wearables data history to be comparable with the new device, or for users who are comparing data with a user who has another type of device, or for groups organizing fitness challenges and health initiatives that can track users by comparing diverse wearables data.
UR - http://www.scopus.com/inward/record.url?scp=85011573430&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85011573430&partnerID=8YFLogxK
U2 - 10.1016/j.compind.2017.01.003
DO - 10.1016/j.compind.2017.01.003
M3 - Article
AN - SCOPUS:85011573430
SN - 0166-3615
VL - 86
SP - 72
EP - 81
JO - Computers in Industry
JF - Computers in Industry
ER -