Model architecture
InjuryShield Home Page
Player Injury Risk Dashboard
Player Statistics Wizard

From Passion to Protection: The Story Behind Injury Shield

Inspiration

Football, the sport we all love and cheer. Whether you're sitting with friends and family, rooting for your favorite team, or screaming at a game-changing play during a fierce rivalry, there's no denying the energy and passion the game brings. Football has a unique ability to connect people, uniting fans from every corner of the world.

But as much as we cheer for the wins, we also feel the sting of injuries that change the course of a game, and sometimes, a player’s future. We may not be football players, but we understand the emotional toll it takes when a key player goes down with an injury, especially one that could have been prevented.

That was the spark that ignited our project: to bring cutting-edge technology into the sport, making injury prediction a reality.

We at Injury Shield protect our players on the field by harnessing the power of technology.

What it does

Injury Shield is an innovative sports and healthcare solution. It leverages advanced machine learning to predict the likelihood of injuries among NFL and college football players. By analyzing extensive historical data—from over 50k data points—our system forecasts potential injuries before they occur with statistically significant accuracy. The platform not only provides risk probabilities for various injury types but also pinpoints probable anatomical locations, enabling coaches and medical teams to tailor training and rehabilitation protocols effectively, while also letting players know what to watch out for in-game. Additionally, our interactive chatbot gathers nuanced information from users, enhancing data accuracy and delivering personalized injury prevention advice.

Importantly, our model’s performance is validated on new, unseen data—demonstrating an R² score of 0.6. This means our system explains 60% of the variance in injury outcomes, making it about 16 times more effective than pure guesswork. Given the inherent uncertainty in predicting injuries (which are probabilistic by nature), achieving an R² of 1 is impossible unless we switched from probabilities to binary outputs. In our context, 0.6 is both statistically significant and practically impactful.

How we built it

Our project is built on a robust framework of data science and software engineering. We developed:

Predictive Injury Model: Using Scikit‑learn and Tensorflow, we created models that evaluate historical and real-time data to generate injury likelihoods. We used statistical methods such as ANOVA and visualization to investigate relationships in the data, and used these relationships to finely craft each layer of our neural network.
Injury Prediction Platform: We created an easy-to-use wizard application for players to input their basic statistics as well as any prior injuries. Then, we used ReactJS to display a colorful and informative dashboard containing the insights generated by AI processes.
Interactive Chatbot Interface: Designed to engage with users and refine predictions by capturing detailed injury histories and current health statuses. Throughout development, we prioritized accuracy and scalability.

Challenges we ran into

Data Acquisition: Before any machine learning could take place, we spent hours researching the perfect dataset to use. And once we had found that dataset, yet more hours were spent carefully cleaning the data, making sure each column and row was in a format that our models could understand.
Machine Learning Hurdles: During training, we encountered common problems in machine learning, such as overfitting and parameter explosion. We solved these challenges by learning about and applying machine learning methods, such as regularization, hyperparameter tuning, and embeddings.
Team Coordination: At first, it was difficult for us to get into the 'groove' of our work as this was the first or second hackathon of everyone on our team. This was the first time many of us had worked on a large-scale group programming project before, so coordination was a bit awkward at first. However, as time went on, everyone eventually found their own niche, and together we accomplished what would have taken any of us alone weeks.

Accomplishments that we're proud of

Despite the challenges, our team achieved several remarkable milestones:

Data Processing: We wrangled multiple datasets (some containing over 100k rows), merging different pieces together to create a comprehensive dataset for our model.
Model Performance: We tested multiple machine learning techniques such as MLPs, Random Forest, and Gradient Boosting (and far more parameters) to train a model that performs 15x better than a naive implementation.
User Experience: We designed a user-friendly interface to present our data, and integrated an LLM for users to ask their own questions.
Real Impact: Our early tests indicate that Injury Shield can be a valuable tool for reducing injury risks and safeguarding athletes' careers.

What we learned

Throughout this journey, we learned the importance of:

Data Quality and Preprocessing: High-quality, well-prepared data is the backbone of any predictive model.
Iterative Model Development: Continuous testing and refinement are essential, especially in complex fields like injury prediction.
User-Centric Design: Thinking about end-user needs—athletes, coaches, trainers, and concerned fans—helped us tailor the system to real-world needs.
Resilience and Adaptability: Overcoming technical and logistical challenges taught us to remain agile and innovative in the face of uncertainty.

Potential Applications

There are many potential applications of our tech! Here are a few we could think of:

A team management tool for coaches to track their players' health while also keeping them safe
A platform for amateur players to improve at the game while being mindful of their health.
A tool for predictive player performance modeling for sports-related wagers

What's next for Injury Shield

Looking ahead, we plan to:

Add new sports: Design different wizards for sports such as tennis, soccer, swimming, etc. and train different models to protect as many athletes as possible.
Expand Data Sources: Incorporate more diverse data, about each player's physical status to enhance prediction accuracy.
Enhance the Chatbot: Improve the interactive interface to gather even richer data and offer more personalized insights.
Integrate Advanced Analytics: Explore deep learning and other advanced techniques to further boost predictive performance.
Expand Impact: Ultimately, we aim to extend our solution to other sports and healthcare areas, transforming how we approach injury prevention across the board.