Small steps. Big impact.

THE FITBIT BLOG

Fitbit Tracker: much more than a pedometer

| 3 Comments

Thanks for the intro Amy!

As Amy mentioned in the last post, my name is Shelten and I’m the Chief Scientist at Fitbit.  I lead our research in sensor architectures and algorithms for health & wellness monitoring and discovery. Since you’re reading this, you probably already know that the Fitbit Tracker is an unobtrusive device that you can wear all day long to track your steps, calories burned, and sleep. In today’s blog post, I’d like to clear up some confusion about how our activity tracking algorithms work.

I. What the Fitbit Tracker does not do

People often refer to the Fitbit Tracker as a “pedometer on steroids”. Illegal drug use aside, I think that’s meant to be a compliment. Thank you. However, that description doesn’t quite give justice to the Tracker.  To understand why, let me first show you what the Fitbit Tracker does NOT do.

Fig. 1 sketches out what a typical pedometer does to estimate your calorie burn. It counts your steps and then multiplies the step count by some value to come up with a calorie value.

II. Question: Why don’t we just convert steps into calories?

Well, it turns out to not be very accurate. People walk and run very differently — some are more efficient at ambulatory activity than others — so a one-size-fits-all conversion of steps to calories burnt won’t work. To see this for yourself, check out Fig. 2 below. If I were to naively say that my additional caloric burn is 8.8 calories per 100 steps (the “Shelten” bar) and then apply that to everyone else, then I’d be undercutting their calories by as much as 43% (as in the case of the “Priscilla” bar).

It gets worse. If I focus on a single user, say James, and look at his data across the week, you can see that a single value doesn’t work well even in that case. Check out Fig. 3. On Wednesday, his value of 6.8 calories per 100 steps is about 46% of the value on Saturday. Presumably this is because on some days James walked more aggressively than others. This implies that even a personalized calibration of a conversion factor from steps to calories won’t do very well if you’re at all interested in knowing how much extra credit you get for going on that jog.

III. How the Fitbit Tracker measures calorie burn

The Fitbit Tracker determines calorie burn by using the *raw* motion data of the user, as obtained from a 3-axis accelerometer. The Tracker looks at the intensity and duration of the accelerometer signal, analyzes patterns, and then determines calorie burn. This general approach to energy expenditure monitoring has roots in the scientific literature dating back to about three decades ago and is still an active area of research in academia and at Fitbit. If you’re interested in learning more about the ongoing state of the field, Staudenmayer, et al. (2009) is a good recent example and van Hees & Ekelund (2009) give a nice overview.

Okay, so what does this all mean? The Fitbit Tracker is much more than a pedometer — it’s also a finely tuned instrument that measures calorie burn. Let’s return to the example of James’s week-long data set, now separated into steps and calories burnt in Fig. 4. We see that by measuring calories and steps as two distinct quantities (rather than just calories derived from steps a la Fig. 1), we can glean some interesting insights into activity. First off, it’s pretty clear that more steps don’t necessarily mean more calories when comparing the values on Thursday to the values on Monday. On Monday, James only walked about 5600 steps but burned 787 calories. On Thursday he walked more (7700 steps) but burned fewer calories (640 calories). If James is trying to lose weight without having to wear out his running shoes, he should try to mimic his behavior on Monday!

IV. References

Staudenmayer, J., Prober, D., Crouter, S., Bassett, D., Freedson, P. (2009) An artificial neural network to estimate physical activity energy expenditure and identify physical activity type from an accelerometer. J Appl Physiol 107: 1300-1307.

van Hees, V.T. and Ekelund, U. (2009) Novel daily energy expenditure estimation by using objective activity type classification: where do we go from here? J Appl Physiol 107: 639-640.

We love data!

| 1 Comment

And we hope you do too – especially data that you can use to help make behavioral change – and live healthier.  And there’s a key member of the Fitbit team we wanted to introduce you to who is our own data expert: Shelten Yuen.  Shelten heads Research at Fitbit and he’s responsible for the “smarts” in the Fitbit tracker. He loves analyzing data and solving complex problems — and you can see it in his previous research in beating heart surgery (with robots, no less) at Harvard University, where he also got his Ph.D., and his research in missile defense at MIT Lincoln Laboratory.

Shelten continues headlong down this path at Fitbit and he’ll be joining us on the blog on a regular basis to share interesting analysis on the data Fitbit collects, and share data trends with the Fitbit community.  Look for his first data analysis piece soon!