Computer scientists working at the San Diego based the University of California have now developed the FitRec which is a recommendation tool that has been powered with the use of deep learning. This tool is now able to perfectly estimate the heart rates of runners during any workout and eventually predict the recommended routes. The team shall also present their completed work at WWW 19 conference to be held between May 13th and May 17th in the San Francisco area.
The researchers involved in the study trained the FitREc over a dataset comprised of 250,000+ workout records which were meant for 1000+ runners. This allowed the computer scientists to create a model which analyzed the past performances in order to predict the speed & heart rate that was given in a specific for specified future routes and workout times.
FitRec is also capable of identifying the crucial features that tend to affect workout performance. This includes things such as a route with hills as well as the fitness level of the user. The tool easily recommends alternate routes that can help the runners achieve a specified target for the heart rate. Additionally, it is perfectly capable of calibrating the short-term predictions which includes telling the runners when they should slow down in order to avoid surpassing the desired heart rate.
The team, however, could develop this tool only partially given the fact that they were actually among the first ones to collect & model the massive dataset for fitness meant for academic research. However, developing FitRec didn’t come as an easy feat given the fact that the fitness-based dataset comes with a massive number of records of a workout but with a minute number for data points for each individual.
Julian McAuley, the professor in Department of Computer Science & Engineering located at UC San Diego mentioned that personalization is actually crucial when it comes to the models for fitness data given the fact that individuals vary immensely from one area to another. This includes heart rate & ability to gel well with various exercises. The researchers also added that the prime challenge in creating this model is the fact that dynamics for heart rates when people exercise is actually very complex and requires sophisticated techniques for modeling them down. In the future, the FitRec could easily be trained to incorporate various other features such as providing runners access to safety-aware routes.