4 Ways to Use the Training Data from Wearable Tech

For special entry to all of our fitness, equipment, journey, and vacation stories, moreover savings on trips, activities, and equipment, signal up for Outdoors+ these days
and help you save fifty per cent.


The central issue that sports scientists are grappling with these days is this: What the heck are we likely to do with all this knowledge? In endurance sports, we’ve progressed from heart amount screens and GPS watches to sophisticated biomechanical assessment, interior oxygen concentrations, and constant glucose measurements, all shown on your wrist then quickly downloaded to your pc. Staff sports have undergone a comparable tech revolution. The ensuing knowledge is interesting and abundant, but is it truly practical?

A new paper in the Worldwide Journal of Sporting activities Physiology and Functionality tackles this issue and presents an intriguing framework for pondering about it, derived from the enterprise analytics literature. The paper will come from Kobe Houtmeyers and Arne Jaspers of KU Leuven in Belgium, along with Pedro Figueiredo of the Portuguese Soccer Federation’s Portugal Soccer School.

Here’s their 4-stage framework for knowledge analytics, offered in purchase of both equally expanding complexity and expanding value to the athlete or mentor:

  • Descriptive: What transpired?
  • Diagnostic: Why did it come about?
  • Predictive: What will come about?
  • Prescriptive: How do we make it come about?

Every single stage builds on the former one particular, which usually means that the descriptive layer is the foundation for almost everything else. Is the knowledge great more than enough? I’m pretty self-assured that a fashionable GPS check out can precisely describe how far and how fast I have run in education, which lets me to move to the following stage and try to diagnose whether or not a great or bad race resulted from education too much, too minor, too tough, too quick, and so on. In contrast, the heart amount knowledge I get from wrist sensors on sports watches is utter rubbish (as verified by evaluating it to knowledge from chest straps). It took me a though to notice that, and any insights I drew from that flawed knowledge would naturally have been meaningless and perhaps harming to my education.

Generating predictions is tougher (especially, as the saying goes, about the long run). Experts in a selection of sports have tried out to use device understanding to comb by huge sets of education knowledge to predict who’s at significant possibility of finding injured. For case in point, a examine released before this year by researchers at the University of Groningen in the Netherlands plugged 7 many years of education and damage knowledge from seventy four aggressive runners into an algorithm that parsed possibility based mostly on either the former 7 days of running (with 10 parameters for each individual day, like the total distance in distinctive education zones, perceived exertion, and duration of cross-education) or the former three months (with 22 parameters per 7 days). The ensuing product, like comparable ones in other sports, was considerably much better than a coin toss at predicting accidents, but not still great more than enough to foundation education selections on.

Prescriptive analytics, the holy grail for sports scientists, is even much more elusive. A easy case in point that does not demand any significant computation is heart-amount variability (HRV), a proxy measure of stress and recovery standing that (as I reviewed in a 2018 posting) has been proposed as a everyday guideline for choosing whether or not to practice tough or quick. Even though the physiology will make sense, I have been skeptical of delegating very important education selections to an algorithm. That is a bogus choice, though, in accordance to Houtmeyers and his colleagues. Prescriptive analytics presents “decision help systems”: the algorithm isn’t replacing the mentor, but is furnishing him or her with a different viewpoint that is not weighed down by the inescapable cognitive biases that afflict human final decision-building.

Apparently, Marco Altini, one particular of the leaders in developing ways to HRV-guided education, posted a Twitter thread a few months in the past in which he reflected on what has transformed in the discipline due to the fact my 2018 posting. Among the insights: the measuring technologies has enhanced, as has know-how about how and when to use it to get the most dependable knowledge. That is important for descriptive utilization. But even great knowledge does not assure great prescriptive advice. According to Altini, scientific studies of HRV-guided education (like this one particular) have moved away from tweaking exercise session options based mostly on the vagaries of that morning’s studying, relying in its place on for a longer time-time period tendencies like running 7-day averages. Even with individuals caveats, I’d however look at HRV as a source of final decision help instead than as a final decision-maker.

Just one of the explanations Houtmeyers’s paper appealed to me is that I put in a bunch of time pondering about these troubles for the duration of my the latest experiment with constant glucose checking. The 4-stage framework allows explain my pondering. It is clear that CGMs provide terrific descriptive knowledge and with some effort, I imagine you can also get some great diagnostic insights. But the sales pitch, as you’d assume, is explicitly concentrated on predictive and prescriptive claims: guiding you on what and when to consume in purchase to optimize efficiency and recovery. Maybe that is achievable, but I’m not still convinced.

In fact, if there’s one particular easy information I choose away from this paper, it is that description and prognosis are not the exact detail as prediction and prescription. The latter does not comply with quickly from the former. As the knowledge sets maintain finding even larger and bigger-good quality, it seems inescapable that we’ll at some point get to the position when device-understanding algorithms can pick up designs and interactions that even very knowledgeable coaches might skip. But that is a huge leap, and knowledge on its own—even “big” data—won’t get us there.

For much more Sweat Science, join me on Twitter and Facebook, signal up for the e mail e-newsletter, and verify out my e book Endure: Mind, Physique, and the Curiously Elastic Boundaries of Human Functionality.