Consumer-grade wearable activity devices such as Fitbits are increasingly being used in research settings to promote physical activity (PA) due to their low-cost and widespread popularity. However, Fitbit-derived measures of activity intensity are consistently reported to be less accurate than intensity estimates obtained from research-grade accelerometers (i.e., ActiGraph). As such, the potential for using a Fitbit to measure PA intensity within research contexts remains limited. This study aims to model ActiGraph-based intensity estimates from the validated Freedson vector magnitude (VM3) algorithm using measures of steps, metabolic equivalents, and intensity levels obtained from Fitbit. Minute-level data collected from 19 subjects, who concurrently wore the ActiGraph GT3X and Fitbit Flex devices for an average of 1.8 weeks, were used to generate the model. After testing several modeling methods, a naïve Bayes classifier was chosen based on the lowest achieved error rate. Overall, the model reduced Fitbit to ActiGraph errors from 19.97% to 16.32%. Moreover, the model reduced misclassification of Fitbit-based estimates of moderate-to-vigorous physical activity (MVPA) by 40%, eliminating a statistically significant difference between MVPA estimates derived from ActiGraph and Fitbit. Study findings support the general utility of the model for measuring MVPA with the Fitbit Flex in place of the more costly ActiGraph GT3X accelerometer for young healthy adults.
- Activity recognition
- error correction
- wearable sensors
ASJC Scopus subject areas
- Computer Science Applications
- Electrical and Electronic Engineering
- Health Information Management