UC Berkeley researchers develop app to modify fitness goals automatically

Mo Zhou/Courtesy
UC Berkeley graduate students Mo Zhou and Yonatan Mintz helped develop an app that uses machine learning to create reasonable, tailored fitness goals.

Related Posts

UC Berkeley researchers have developed an exercise app designed to keep consumers motivated by establishing personalized, dynamic and realistic daily step goals using technology that determines the user’s physical activity.

CAL Fit, which is similar to exercise apps such as Fitbit and Google Fit, tracks the number of steps a user takes daily and sets future goals based on current physical activity levels. Unlike its more established counterparts, however, CAL Fit is able to modify an individual’s future fitness goals through a machine learning algorithm, which allows the app to gauge an individual’s motivation to exercise based on data from their past steps and goals.

The app was designed by a team of two graduate students, three UC Berkeley faculty members, two UCSF faculty members and three undergraduate students who helped with the experiments.

The individually tailored goals set CAL Fit apart from other exercise apps, according to Yonatan Mintz, a campus graduate student who developed the algorithms that guide the app.

“For the first time, we’re actually using data that people generate themselves to fit their preferences and their needs to get whatever health care outcomes they want to get, as opposed to making blanket statements about populations,” Mintz said.

According to campus graduate student Mo Zhou, who helped lead the development of CAL Fit, the goal is to create a fitness app that will help consumers achieve their fitness goals in the long run.

“From personal experience, when I want to do activity one day and then I download (other fitness) apps, it does not motivate me to continuously do activity — I always give up within a week,” Zhou said. “One of the issues we found was that goal-setting (in other apps) is not dynamic and personalized. That’s why we think there is a lot of room there, and we can use a smarter algorithm to improve this experience.”

As for how this goal will be achieved, the machine learning algorithm collects baseline data that are then used to create a model for each subject. From there, the app determines how likely a person is to achieve future goals based on their history of meeting past goals.

“I would use (CAL Fit),” said campus freshman Katerine Perez, who was exercising at the Recreational Sports Facility. “I currently have the Fitbit app, and you can change the goals, but you have to make your own decisions, which is hard for someone who doesn’t know about exercise and nutrition. You have to go out of your way to do your own research.

CAL Fit is not yet available for purchase on Apple and Android phones, according to Zhou. The team is still adding updates to the app, including a notification system to keep users motivated, according to Anil Aswani, assistant professor of industrial engineering and operations research and one of the lead project researchers.

“We do think that the motivational messages will encourage people to increase their activities,” Aswani said. “We believe personalizing these messages will increase their effectiveness.”

Contact Shelby Mayes at [email protected] and follow her on Twitter at @shelbymayesdc.