Tricking your fitness tracker into logging a workout when you are in fact just laying on the couch seems like a fairly futile exercise, but there's more to the equation than just fooling yourself. Insurers and health care providers are increasingly relying on tracking data to offer incentives, reduced premiums and keep tabs on clients behavior. This is cause for concern for one team of US researchers, which has developed an activity tracking smartphone app that can better distinguish between real and imitated physical movement.

Fitness trackers such as Fitbit and Jawbone that monitor things like heart rate and the amount of steps taken have become a useful tool for health insurers looking for a competitive advantage. One example is New York's Oscar Health – last year the company decided to ship customers complimentary fitness trackers and pledged to reward those clocking up enough steps with Amazon gift cards.

Australian company MLC offers members 10 percent discounts on life insurance if they meet certain physical criteria, as monitored through a Basis Peak fitness tracker. And these connected devices are making it easier than ever to share fitness data with doctors, a situation where misinformation would pose even more serious problems.

"As health care providers and insurance companies rely more on activity trackers, there is an imminent need to make these systems smarter against deceptive behavior," says Sohrab Saeb, a postdoctoral fellow at Northwestern University Feinberg School of Medicine. "We've shown how to train systems to make sure data is authentic."

Saeb led a team of researchers to develop a fitness tracking system that can better detect when the subject is cheating. This began with 14 subjects, who were asked to try and trick an Android smartphone activity app into logging physical tasks they weren't actually carrying out, such as shaking the phone while seated, or swinging their arms back and forth to mimic walking.

If they succeeded, the team used motion data captured from the device's accelerometer and gyroscope to retrain the app to recognize their trickery. This process was repeated up to six times to account for the varied methods of cheating, until the subjects were unable to deceive the system.

The team says regular activity classifiers predict true activity with around 38 percent accuracy, while their solution based on data gleaned from the cheating exercises resulted in 84 percent accuracy. It claims that learning the cheating tactics of one person helps to detect the tactics of others, and the technology can therefore be generalized to make for more effective activity classification overall.

"Very few studies have tried to make activity tracking recognition robust against cheating," says senior author Konrad Kording, a scientist at Rehabilitation Institute of Chicago (RIC) and member of the research team. "This technology could have broad implications for companies that make activity trackers and insurance companies alike as they seek to more reliably record movement."

While smartphones were used in this study, the researchers say the same technology can be applied to fitness tracking bracelets and other wearables as well.

The research was published in the journal PLOS One.