NFL teams experience highly varied success rates for preventing on-field injuries using off-season workout routines. While there are various theories on the ability of workout routines to prevent injuries, there is no data-driven study quantifying the ability of specific workouts to truly prevent injuries.
What it does
We have an easy-to-use website for high school football teams across the country, where coaches or players can input their own workout routines and injury history. This data goes into our data system, where we use a neural network model to predict the likelihood of injury. Every new data point goes back into the system to continuously train our model.
How we built it
We built the web application through the use of the flask python web framework. We connected the html templates to provide to an easy to use user interface and styled the pages with css. We then machine learned the user data to create a prediction model in order to determine if a certain workout routine would result in an injury.
Challenges we ran into
Because this aggregated data is not available anywhere on the market, we could not train an accurate model for predicting workout safety. To work around this, we found average injury rates of high school football and created our own randomized, slightly biased data, to cross-validate the accuracy of our model. With real-world data, our model would be much more accurate, with real correlations, rather than the data we have currently.
Accomplishments that we're proud of
We are proud of the prediction model that we created through the TensorFlow library. We are proud to deliver an appealing web user interface that consumers can seamlessly use to to predict their injuries.
What's next for Crowdsourcing Healthier Workouts
We are going to push our website and crowdsourcing system to market, to actual high school football teams across the market. Once we have a stream of continuous data, we will have a truly accurate system for modeling the safety of workout routines.