Scientists have built a system based on deep learning to recommend personalised workout sessions to help users reach their ideal heart rate during running


deep learning tool to help reach ideal heart rate (Credit: pexels)

A new fitness tracking tool powered by deep learning can recommend workout moves to predict and develop an ideal heart rate during running and working out.

Designed by computer scientists at the University of California in San Diego, FitRec tested on a data set of more than 250,000 workout records for more than a thousand runners.

This enables the device to analyse past performances to predict speed and heart rate for future workout times and routes, which is “slightly harder or will put the user’s heart rate within a certain range”.

Julian McAuley, a professor in the department of computer science and engineering at University of California, said: “Personalisation is crucial in models of fitness data because individuals vary widely in many areas, including heart rate and ability to adapt to different exercises.


How does the FitRec learn and predict workout activities?

From fitness wearable trackers to heart rate monitors, making workout routines more productive is an objective every company in the healthtech industry is attempting to achieve.

Therefore, incorporating AI and deep learning in healthtechs to give users a more personalised experience to enhance engagement and performance is becoming a popular trend right now.

ideal heart rate during running
FitRec’s AI device enables more productive workout sessions (Credit: Pixabay)

Mr McAuley acknowledges the progress in existing tracking tools, but is aiming to go beyond the limits of just tracking.

He said: “We’re hoping to build something that is not just storing a log of what exercises you’ve already done, but can recommend new routines that are personalised based on your past activities.

“For example, we could recommend a running route while the user is traveling that’s similar [in terms of your heart rate profile] to the one they normally run at home, or we could recommend a route that’s slightly harder, or which will put your heart rate within a certain range.

“In the future, we’d hope to use this technology to recommend entire sequences of exercises that could help achieve personalised fitness goals.”

To acquire these advanced features, the team of researchers used a type of deep learning architecture called Recurrent Neural Networks, which they adapted to learn the dynamics and relationships between different activities, routes and heart rate profiles.

Creating FitRec was a challenging task for the team, but the most difficult factors included personalising the system due to the limited amounts of data per user.

Mr McAuley explained: “Heart rate dynamics are complex and thus require lots of data to capture with a deep learning model.

“Although we have large amounts of such data, there’s only a small amount per individual user based on which the model has to tailor its predictions.

“Also, dealing with data quality issues – for example if a user forgets to turn off their device when they return to their car.

“Such issues can quickly confuse a predictive model and had to be carefully filtered.”


Who would benefit most from using FitRec?

As the deep learning tool’s main advantage is capable of recommending alternate routes for runners to either speed up or avoid exceeding their ideal heart rate during running, FitRec is useful for anyone wanting more value out of their time working out.

“The people most likely to benefit from this technology are people looking to add more variety to their exercise routines,” explained Mr McAuley.

He also explained the recommendation device will be beneficial for those working to “tune” their exercise routines by reaching a target peak heart rate or a pace that will keep their heart rate within a certain range, and those who want real-time guidance.

He added: “In the future we’d hope to extend the technology to help users achieve long term fitness goals through personalised exercise plans.”

Researchers also say the tool could also be applied to more complex recommendation routes, for example safety-aware routes.

The team will present their work at the World Wide Web 2019 conference held on May 13 to 17 in San Francisco, an international conference on the topic of the internet.