FitRec is a tool developed by the scientists at the University of California San Diego. As a recommendation tool powered by deep learning, it can estimate the heart rates of runners during a workout and predict as well as recommend routes.
The scientists prepared FitRec on a dataset of more than 240,000 workout records for more than 1,000 runners. This training allowed computer scientists to build a model that analyzed past performance to predict speed and heart rate given specific future workout times and routes.
Apart from that, the tool is capable of identifying essential features that affect workout performance like whether a route has hills and the user's level of fitness. FitRec can recommend alternate routes for runners who want to achieve a specific target heart rate. Also, it is capable of making short-term predictions including informing runners when to slow down to avoid exceeding their desired maximum heart rate.
The scientists were able to develop the tool partially because they were among the first to collect and model a massive fitness dataset for academic research. But it was not an easy feat to develop FitRec as the fitness dataset has a vast number of workout records but only a small number of data points per individual.
A professor in the Department of Computer Science and Engineering at UC San Diego, Julian McAuley, said that personalization is critical in models of fitness data because individuals vary widely in many areas, including heart rate and ability to different exercises. Julian added that the main challenge in building this type of models is that the dynamics of heart rates as people exercise are incredibly complicated, and it requires sophisticated techniques to model.
For them to develop a useful model, computer scientists needed a tool that uses all of the data to learn but at the same time can learn personalized dynamics from a small number of data points per user. Then, they used a deep learning architecture called short-term memory networks (or LSTM), which they adapted to capture the individual dynamics behaviors of each user in the dataset.
To validate the predictions from FitRec, the scientists compared them with existing workout records that were not part of the training dataset.
It will be possible to train FitRec in the future to include other data such as the way users' fitness level evolve to make its predictions. The tool could also be applied to more complex recommendation routes, such as safety-aware routes.
Ultimately, the researchers need to have access to more detailed fitness tracking data and tackle the various data quality issues for the tool to be used in commercial fitness apps.
