IEOR Graduate Students Create New Exercise App using Machine Learning
March 6, 2018

IEOR graduate students Mo Zhou and Yonatan Mintz are addressing exercise commitment issues with a new app that uses machine learning to dynamically adjust goals on a daily basis.

Each New Year around 20% of Americans vow to lose weight, eat healthier, or exercise more. However, for many, this New Year's resolution is one of the most difficult to keep. Work, school, and daily stress can make it difficult to stay on track to meeting fitness goals. Setting static goals (e.g. "walk 10,000 steps daily") can negatively affect morale when the goals are consistently unmet and may even reduce chances of getting fit.

“Some exercise apps automate exercise goals [...] but they can’t adapt to an individual's success [or lack of thereof]. As a result, the goals can get out of sync with the actual performance,” said IEOR Professor Anil Aswani who oversaw the research. 

Some exercise apps do have goals that adapt based on the user's results, but this usually happens through the use of human coaches. However, because human support is typically more expensive than when using an automated system, the cost of the new app is significantly lower than others on the market that utilize human input. 

The app was tested in a controlled experiment with 64 participants. While all participants did not perform as well as they originally hoped, the results were promising as the participants that used the new app logged 10% more steps on average than the control group. Zhou and Mintz hope to take America into a healthier future one step at a time. 

Read the full story at Berkeley Engineering

Also covered in the Daily Californian